xref: /openbmc/qemu/hw/i386/intel_iommu.c (revision 745a4f5e)
1 /*
2  * QEMU emulation of an Intel IOMMU (VT-d)
3  *   (DMA Remapping device)
4  *
5  * Copyright (C) 2013 Knut Omang, Oracle <knut.omang@oracle.com>
6  * Copyright (C) 2014 Le Tan, <tamlokveer@gmail.com>
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12 
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17 
18  * You should have received a copy of the GNU General Public License along
19  * with this program; if not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #include "qemu/osdep.h"
23 #include "qemu/error-report.h"
24 #include "qapi/error.h"
25 #include "hw/sysbus.h"
26 #include "exec/address-spaces.h"
27 #include "intel_iommu_internal.h"
28 #include "hw/pci/pci.h"
29 #include "hw/pci/pci_bus.h"
30 #include "hw/i386/pc.h"
31 #include "hw/i386/apic-msidef.h"
32 #include "hw/boards.h"
33 #include "hw/i386/x86-iommu.h"
34 #include "hw/pci-host/q35.h"
35 #include "sysemu/kvm.h"
36 #include "hw/i386/apic_internal.h"
37 #include "kvm_i386.h"
38 #include "trace.h"
39 
40 static void vtd_address_space_refresh_all(IntelIOMMUState *s);
41 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n);
42 
43 static void vtd_define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val,
44                             uint64_t wmask, uint64_t w1cmask)
45 {
46     stq_le_p(&s->csr[addr], val);
47     stq_le_p(&s->wmask[addr], wmask);
48     stq_le_p(&s->w1cmask[addr], w1cmask);
49 }
50 
51 static void vtd_define_quad_wo(IntelIOMMUState *s, hwaddr addr, uint64_t mask)
52 {
53     stq_le_p(&s->womask[addr], mask);
54 }
55 
56 static void vtd_define_long(IntelIOMMUState *s, hwaddr addr, uint32_t val,
57                             uint32_t wmask, uint32_t w1cmask)
58 {
59     stl_le_p(&s->csr[addr], val);
60     stl_le_p(&s->wmask[addr], wmask);
61     stl_le_p(&s->w1cmask[addr], w1cmask);
62 }
63 
64 static void vtd_define_long_wo(IntelIOMMUState *s, hwaddr addr, uint32_t mask)
65 {
66     stl_le_p(&s->womask[addr], mask);
67 }
68 
69 /* "External" get/set operations */
70 static void vtd_set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
71 {
72     uint64_t oldval = ldq_le_p(&s->csr[addr]);
73     uint64_t wmask = ldq_le_p(&s->wmask[addr]);
74     uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
75     stq_le_p(&s->csr[addr],
76              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
77 }
78 
79 static void vtd_set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
80 {
81     uint32_t oldval = ldl_le_p(&s->csr[addr]);
82     uint32_t wmask = ldl_le_p(&s->wmask[addr]);
83     uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
84     stl_le_p(&s->csr[addr],
85              ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
86 }
87 
88 static uint64_t vtd_get_quad(IntelIOMMUState *s, hwaddr addr)
89 {
90     uint64_t val = ldq_le_p(&s->csr[addr]);
91     uint64_t womask = ldq_le_p(&s->womask[addr]);
92     return val & ~womask;
93 }
94 
95 static uint32_t vtd_get_long(IntelIOMMUState *s, hwaddr addr)
96 {
97     uint32_t val = ldl_le_p(&s->csr[addr]);
98     uint32_t womask = ldl_le_p(&s->womask[addr]);
99     return val & ~womask;
100 }
101 
102 /* "Internal" get/set operations */
103 static uint64_t vtd_get_quad_raw(IntelIOMMUState *s, hwaddr addr)
104 {
105     return ldq_le_p(&s->csr[addr]);
106 }
107 
108 static uint32_t vtd_get_long_raw(IntelIOMMUState *s, hwaddr addr)
109 {
110     return ldl_le_p(&s->csr[addr]);
111 }
112 
113 static void vtd_set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t val)
114 {
115     stq_le_p(&s->csr[addr], val);
116 }
117 
118 static uint32_t vtd_set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
119                                         uint32_t clear, uint32_t mask)
120 {
121     uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
122     stl_le_p(&s->csr[addr], new_val);
123     return new_val;
124 }
125 
126 static uint64_t vtd_set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
127                                         uint64_t clear, uint64_t mask)
128 {
129     uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
130     stq_le_p(&s->csr[addr], new_val);
131     return new_val;
132 }
133 
134 static inline void vtd_iommu_lock(IntelIOMMUState *s)
135 {
136     qemu_mutex_lock(&s->iommu_lock);
137 }
138 
139 static inline void vtd_iommu_unlock(IntelIOMMUState *s)
140 {
141     qemu_mutex_unlock(&s->iommu_lock);
142 }
143 
144 /* Whether the address space needs to notify new mappings */
145 static inline gboolean vtd_as_has_map_notifier(VTDAddressSpace *as)
146 {
147     return as->notifier_flags & IOMMU_NOTIFIER_MAP;
148 }
149 
150 /* GHashTable functions */
151 static gboolean vtd_uint64_equal(gconstpointer v1, gconstpointer v2)
152 {
153     return *((const uint64_t *)v1) == *((const uint64_t *)v2);
154 }
155 
156 static guint vtd_uint64_hash(gconstpointer v)
157 {
158     return (guint)*(const uint64_t *)v;
159 }
160 
161 static gboolean vtd_hash_remove_by_domain(gpointer key, gpointer value,
162                                           gpointer user_data)
163 {
164     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
165     uint16_t domain_id = *(uint16_t *)user_data;
166     return entry->domain_id == domain_id;
167 }
168 
169 /* The shift of an addr for a certain level of paging structure */
170 static inline uint32_t vtd_slpt_level_shift(uint32_t level)
171 {
172     assert(level != 0);
173     return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
174 }
175 
176 static inline uint64_t vtd_slpt_level_page_mask(uint32_t level)
177 {
178     return ~((1ULL << vtd_slpt_level_shift(level)) - 1);
179 }
180 
181 static gboolean vtd_hash_remove_by_page(gpointer key, gpointer value,
182                                         gpointer user_data)
183 {
184     VTDIOTLBEntry *entry = (VTDIOTLBEntry *)value;
185     VTDIOTLBPageInvInfo *info = (VTDIOTLBPageInvInfo *)user_data;
186     uint64_t gfn = (info->addr >> VTD_PAGE_SHIFT_4K) & info->mask;
187     uint64_t gfn_tlb = (info->addr & entry->mask) >> VTD_PAGE_SHIFT_4K;
188     return (entry->domain_id == info->domain_id) &&
189             (((entry->gfn & info->mask) == gfn) ||
190              (entry->gfn == gfn_tlb));
191 }
192 
193 /* Reset all the gen of VTDAddressSpace to zero and set the gen of
194  * IntelIOMMUState to 1.  Must be called with IOMMU lock held.
195  */
196 static void vtd_reset_context_cache_locked(IntelIOMMUState *s)
197 {
198     VTDAddressSpace *vtd_as;
199     VTDBus *vtd_bus;
200     GHashTableIter bus_it;
201     uint32_t devfn_it;
202 
203     trace_vtd_context_cache_reset();
204 
205     g_hash_table_iter_init(&bus_it, s->vtd_as_by_busptr);
206 
207     while (g_hash_table_iter_next (&bus_it, NULL, (void**)&vtd_bus)) {
208         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
209             vtd_as = vtd_bus->dev_as[devfn_it];
210             if (!vtd_as) {
211                 continue;
212             }
213             vtd_as->context_cache_entry.context_cache_gen = 0;
214         }
215     }
216     s->context_cache_gen = 1;
217 }
218 
219 /* Must be called with IOMMU lock held. */
220 static void vtd_reset_iotlb_locked(IntelIOMMUState *s)
221 {
222     assert(s->iotlb);
223     g_hash_table_remove_all(s->iotlb);
224 }
225 
226 static void vtd_reset_iotlb(IntelIOMMUState *s)
227 {
228     vtd_iommu_lock(s);
229     vtd_reset_iotlb_locked(s);
230     vtd_iommu_unlock(s);
231 }
232 
233 static void vtd_reset_caches(IntelIOMMUState *s)
234 {
235     vtd_iommu_lock(s);
236     vtd_reset_iotlb_locked(s);
237     vtd_reset_context_cache_locked(s);
238     vtd_iommu_unlock(s);
239 }
240 
241 static uint64_t vtd_get_iotlb_key(uint64_t gfn, uint16_t source_id,
242                                   uint32_t level)
243 {
244     return gfn | ((uint64_t)(source_id) << VTD_IOTLB_SID_SHIFT) |
245            ((uint64_t)(level) << VTD_IOTLB_LVL_SHIFT);
246 }
247 
248 static uint64_t vtd_get_iotlb_gfn(hwaddr addr, uint32_t level)
249 {
250     return (addr & vtd_slpt_level_page_mask(level)) >> VTD_PAGE_SHIFT_4K;
251 }
252 
253 /* Must be called with IOMMU lock held */
254 static VTDIOTLBEntry *vtd_lookup_iotlb(IntelIOMMUState *s, uint16_t source_id,
255                                        hwaddr addr)
256 {
257     VTDIOTLBEntry *entry;
258     uint64_t key;
259     int level;
260 
261     for (level = VTD_SL_PT_LEVEL; level < VTD_SL_PML4_LEVEL; level++) {
262         key = vtd_get_iotlb_key(vtd_get_iotlb_gfn(addr, level),
263                                 source_id, level);
264         entry = g_hash_table_lookup(s->iotlb, &key);
265         if (entry) {
266             goto out;
267         }
268     }
269 
270 out:
271     return entry;
272 }
273 
274 /* Must be with IOMMU lock held */
275 static void vtd_update_iotlb(IntelIOMMUState *s, uint16_t source_id,
276                              uint16_t domain_id, hwaddr addr, uint64_t slpte,
277                              uint8_t access_flags, uint32_t level)
278 {
279     VTDIOTLBEntry *entry = g_malloc(sizeof(*entry));
280     uint64_t *key = g_malloc(sizeof(*key));
281     uint64_t gfn = vtd_get_iotlb_gfn(addr, level);
282 
283     trace_vtd_iotlb_page_update(source_id, addr, slpte, domain_id);
284     if (g_hash_table_size(s->iotlb) >= VTD_IOTLB_MAX_SIZE) {
285         trace_vtd_iotlb_reset("iotlb exceeds size limit");
286         vtd_reset_iotlb_locked(s);
287     }
288 
289     entry->gfn = gfn;
290     entry->domain_id = domain_id;
291     entry->slpte = slpte;
292     entry->access_flags = access_flags;
293     entry->mask = vtd_slpt_level_page_mask(level);
294     *key = vtd_get_iotlb_key(gfn, source_id, level);
295     g_hash_table_replace(s->iotlb, key, entry);
296 }
297 
298 /* Given the reg addr of both the message data and address, generate an
299  * interrupt via MSI.
300  */
301 static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
302                                    hwaddr mesg_data_reg)
303 {
304     MSIMessage msi;
305 
306     assert(mesg_data_reg < DMAR_REG_SIZE);
307     assert(mesg_addr_reg < DMAR_REG_SIZE);
308 
309     msi.address = vtd_get_long_raw(s, mesg_addr_reg);
310     msi.data = vtd_get_long_raw(s, mesg_data_reg);
311 
312     trace_vtd_irq_generate(msi.address, msi.data);
313 
314     apic_get_class()->send_msi(&msi);
315 }
316 
317 /* Generate a fault event to software via MSI if conditions are met.
318  * Notice that the value of FSTS_REG being passed to it should be the one
319  * before any update.
320  */
321 static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
322 {
323     if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
324         pre_fsts & VTD_FSTS_IQE) {
325         error_report_once("There are previous interrupt conditions "
326                           "to be serviced by software, fault event "
327                           "is not generated");
328         return;
329     }
330     vtd_set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
331     if (vtd_get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
332         error_report_once("Interrupt Mask set, irq is not generated");
333     } else {
334         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
335         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
336     }
337 }
338 
339 /* Check if the Fault (F) field of the Fault Recording Register referenced by
340  * @index is Set.
341  */
342 static bool vtd_is_frcd_set(IntelIOMMUState *s, uint16_t index)
343 {
344     /* Each reg is 128-bit */
345     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
346     addr += 8; /* Access the high 64-bit half */
347 
348     assert(index < DMAR_FRCD_REG_NR);
349 
350     return vtd_get_quad_raw(s, addr) & VTD_FRCD_F;
351 }
352 
353 /* Update the PPF field of Fault Status Register.
354  * Should be called whenever change the F field of any fault recording
355  * registers.
356  */
357 static void vtd_update_fsts_ppf(IntelIOMMUState *s)
358 {
359     uint32_t i;
360     uint32_t ppf_mask = 0;
361 
362     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
363         if (vtd_is_frcd_set(s, i)) {
364             ppf_mask = VTD_FSTS_PPF;
365             break;
366         }
367     }
368     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
369     trace_vtd_fsts_ppf(!!ppf_mask);
370 }
371 
372 static void vtd_set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t index)
373 {
374     /* Each reg is 128-bit */
375     hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
376     addr += 8; /* Access the high 64-bit half */
377 
378     assert(index < DMAR_FRCD_REG_NR);
379 
380     vtd_set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
381     vtd_update_fsts_ppf(s);
382 }
383 
384 /* Must not update F field now, should be done later */
385 static void vtd_record_frcd(IntelIOMMUState *s, uint16_t index,
386                             uint16_t source_id, hwaddr addr,
387                             VTDFaultReason fault, bool is_write)
388 {
389     uint64_t hi = 0, lo;
390     hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
391 
392     assert(index < DMAR_FRCD_REG_NR);
393 
394     lo = VTD_FRCD_FI(addr);
395     hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
396     if (!is_write) {
397         hi |= VTD_FRCD_T;
398     }
399     vtd_set_quad_raw(s, frcd_reg_addr, lo);
400     vtd_set_quad_raw(s, frcd_reg_addr + 8, hi);
401 
402     trace_vtd_frr_new(index, hi, lo);
403 }
404 
405 /* Try to collapse multiple pending faults from the same requester */
406 static bool vtd_try_collapse_fault(IntelIOMMUState *s, uint16_t source_id)
407 {
408     uint32_t i;
409     uint64_t frcd_reg;
410     hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
411 
412     for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
413         frcd_reg = vtd_get_quad_raw(s, addr);
414         if ((frcd_reg & VTD_FRCD_F) &&
415             ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
416             return true;
417         }
418         addr += 16; /* 128-bit for each */
419     }
420     return false;
421 }
422 
423 /* Log and report an DMAR (address translation) fault to software */
424 static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
425                                   hwaddr addr, VTDFaultReason fault,
426                                   bool is_write)
427 {
428     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
429 
430     assert(fault < VTD_FR_MAX);
431 
432     if (fault == VTD_FR_RESERVED_ERR) {
433         /* This is not a normal fault reason case. Drop it. */
434         return;
435     }
436 
437     trace_vtd_dmar_fault(source_id, fault, addr, is_write);
438 
439     if (fsts_reg & VTD_FSTS_PFO) {
440         error_report_once("New fault is not recorded due to "
441                           "Primary Fault Overflow");
442         return;
443     }
444 
445     if (vtd_try_collapse_fault(s, source_id)) {
446         error_report_once("New fault is not recorded due to "
447                           "compression of faults");
448         return;
449     }
450 
451     if (vtd_is_frcd_set(s, s->next_frcd_reg)) {
452         error_report_once("Next Fault Recording Reg is used, "
453                           "new fault is not recorded, set PFO field");
454         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
455         return;
456     }
457 
458     vtd_record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
459 
460     if (fsts_reg & VTD_FSTS_PPF) {
461         error_report_once("There are pending faults already, "
462                           "fault event is not generated");
463         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg);
464         s->next_frcd_reg++;
465         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
466             s->next_frcd_reg = 0;
467         }
468     } else {
469         vtd_set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
470                                 VTD_FSTS_FRI(s->next_frcd_reg));
471         vtd_set_frcd_and_update_ppf(s, s->next_frcd_reg); /* Will set PPF */
472         s->next_frcd_reg++;
473         if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
474             s->next_frcd_reg = 0;
475         }
476         /* This case actually cause the PPF to be Set.
477          * So generate fault event (interrupt).
478          */
479          vtd_generate_fault_event(s, fsts_reg);
480     }
481 }
482 
483 /* Handle Invalidation Queue Errors of queued invalidation interface error
484  * conditions.
485  */
486 static void vtd_handle_inv_queue_error(IntelIOMMUState *s)
487 {
488     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
489 
490     vtd_set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_IQE);
491     vtd_generate_fault_event(s, fsts_reg);
492 }
493 
494 /* Set the IWC field and try to generate an invalidation completion interrupt */
495 static void vtd_generate_completion_event(IntelIOMMUState *s)
496 {
497     if (vtd_get_long_raw(s, DMAR_ICS_REG) & VTD_ICS_IWC) {
498         trace_vtd_inv_desc_wait_irq("One pending, skip current");
499         return;
500     }
501     vtd_set_clear_mask_long(s, DMAR_ICS_REG, 0, VTD_ICS_IWC);
502     vtd_set_clear_mask_long(s, DMAR_IECTL_REG, 0, VTD_IECTL_IP);
503     if (vtd_get_long_raw(s, DMAR_IECTL_REG) & VTD_IECTL_IM) {
504         trace_vtd_inv_desc_wait_irq("IM in IECTL_REG is set, "
505                                     "new event not generated");
506         return;
507     } else {
508         /* Generate the interrupt event */
509         trace_vtd_inv_desc_wait_irq("Generating complete event");
510         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
511         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
512     }
513 }
514 
515 static inline bool vtd_root_entry_present(VTDRootEntry *root)
516 {
517     return root->val & VTD_ROOT_ENTRY_P;
518 }
519 
520 static int vtd_get_root_entry(IntelIOMMUState *s, uint8_t index,
521                               VTDRootEntry *re)
522 {
523     dma_addr_t addr;
524 
525     addr = s->root + index * sizeof(*re);
526     if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
527         trace_vtd_re_invalid(re->rsvd, re->val);
528         re->val = 0;
529         return -VTD_FR_ROOT_TABLE_INV;
530     }
531     re->val = le64_to_cpu(re->val);
532     return 0;
533 }
534 
535 static inline bool vtd_ce_present(VTDContextEntry *context)
536 {
537     return context->lo & VTD_CONTEXT_ENTRY_P;
538 }
539 
540 static int vtd_get_context_entry_from_root(VTDRootEntry *root, uint8_t index,
541                                            VTDContextEntry *ce)
542 {
543     dma_addr_t addr;
544 
545     /* we have checked that root entry is present */
546     addr = (root->val & VTD_ROOT_ENTRY_CTP) + index * sizeof(*ce);
547     if (dma_memory_read(&address_space_memory, addr, ce, sizeof(*ce))) {
548         trace_vtd_re_invalid(root->rsvd, root->val);
549         return -VTD_FR_CONTEXT_TABLE_INV;
550     }
551     ce->lo = le64_to_cpu(ce->lo);
552     ce->hi = le64_to_cpu(ce->hi);
553     return 0;
554 }
555 
556 static inline dma_addr_t vtd_ce_get_slpt_base(VTDContextEntry *ce)
557 {
558     return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
559 }
560 
561 static inline uint64_t vtd_get_slpte_addr(uint64_t slpte, uint8_t aw)
562 {
563     return slpte & VTD_SL_PT_BASE_ADDR_MASK(aw);
564 }
565 
566 /* Whether the pte indicates the address of the page frame */
567 static inline bool vtd_is_last_slpte(uint64_t slpte, uint32_t level)
568 {
569     return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
570 }
571 
572 /* Get the content of a spte located in @base_addr[@index] */
573 static uint64_t vtd_get_slpte(dma_addr_t base_addr, uint32_t index)
574 {
575     uint64_t slpte;
576 
577     assert(index < VTD_SL_PT_ENTRY_NR);
578 
579     if (dma_memory_read(&address_space_memory,
580                         base_addr + index * sizeof(slpte), &slpte,
581                         sizeof(slpte))) {
582         slpte = (uint64_t)-1;
583         return slpte;
584     }
585     slpte = le64_to_cpu(slpte);
586     return slpte;
587 }
588 
589 /* Given an iova and the level of paging structure, return the offset
590  * of current level.
591  */
592 static inline uint32_t vtd_iova_level_offset(uint64_t iova, uint32_t level)
593 {
594     return (iova >> vtd_slpt_level_shift(level)) &
595             ((1ULL << VTD_SL_LEVEL_BITS) - 1);
596 }
597 
598 /* Check Capability Register to see if the @level of page-table is supported */
599 static inline bool vtd_is_level_supported(IntelIOMMUState *s, uint32_t level)
600 {
601     return VTD_CAP_SAGAW_MASK & s->cap &
602            (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
603 }
604 
605 /* Get the page-table level that hardware should use for the second-level
606  * page-table walk from the Address Width field of context-entry.
607  */
608 static inline uint32_t vtd_ce_get_level(VTDContextEntry *ce)
609 {
610     return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
611 }
612 
613 static inline uint32_t vtd_ce_get_agaw(VTDContextEntry *ce)
614 {
615     return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
616 }
617 
618 static inline uint32_t vtd_ce_get_type(VTDContextEntry *ce)
619 {
620     return ce->lo & VTD_CONTEXT_ENTRY_TT;
621 }
622 
623 /* Return true if check passed, otherwise false */
624 static inline bool vtd_ce_type_check(X86IOMMUState *x86_iommu,
625                                      VTDContextEntry *ce)
626 {
627     switch (vtd_ce_get_type(ce)) {
628     case VTD_CONTEXT_TT_MULTI_LEVEL:
629         /* Always supported */
630         break;
631     case VTD_CONTEXT_TT_DEV_IOTLB:
632         if (!x86_iommu->dt_supported) {
633             return false;
634         }
635         break;
636     case VTD_CONTEXT_TT_PASS_THROUGH:
637         if (!x86_iommu->pt_supported) {
638             return false;
639         }
640         break;
641     default:
642         /* Unknwon type */
643         return false;
644     }
645     return true;
646 }
647 
648 static inline uint64_t vtd_iova_limit(VTDContextEntry *ce, uint8_t aw)
649 {
650     uint32_t ce_agaw = vtd_ce_get_agaw(ce);
651     return 1ULL << MIN(ce_agaw, aw);
652 }
653 
654 /* Return true if IOVA passes range check, otherwise false. */
655 static inline bool vtd_iova_range_check(uint64_t iova, VTDContextEntry *ce,
656                                         uint8_t aw)
657 {
658     /*
659      * Check if @iova is above 2^X-1, where X is the minimum of MGAW
660      * in CAP_REG and AW in context-entry.
661      */
662     return !(iova & ~(vtd_iova_limit(ce, aw) - 1));
663 }
664 
665 /*
666  * Rsvd field masks for spte:
667  *     Index [1] to [4] 4k pages
668  *     Index [5] to [8] large pages
669  */
670 static uint64_t vtd_paging_entry_rsvd_field[9];
671 
672 static bool vtd_slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
673 {
674     if (slpte & VTD_SL_PT_PAGE_SIZE_MASK) {
675         /* Maybe large page */
676         return slpte & vtd_paging_entry_rsvd_field[level + 4];
677     } else {
678         return slpte & vtd_paging_entry_rsvd_field[level];
679     }
680 }
681 
682 /* Find the VTD address space associated with a given bus number */
683 static VTDBus *vtd_find_as_from_bus_num(IntelIOMMUState *s, uint8_t bus_num)
684 {
685     VTDBus *vtd_bus = s->vtd_as_by_bus_num[bus_num];
686     if (!vtd_bus) {
687         /*
688          * Iterate over the registered buses to find the one which
689          * currently hold this bus number, and update the bus_num
690          * lookup table:
691          */
692         GHashTableIter iter;
693 
694         g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
695         while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
696             if (pci_bus_num(vtd_bus->bus) == bus_num) {
697                 s->vtd_as_by_bus_num[bus_num] = vtd_bus;
698                 return vtd_bus;
699             }
700         }
701     }
702     return vtd_bus;
703 }
704 
705 /* Given the @iova, get relevant @slptep. @slpte_level will be the last level
706  * of the translation, can be used for deciding the size of large page.
707  */
708 static int vtd_iova_to_slpte(VTDContextEntry *ce, uint64_t iova, bool is_write,
709                              uint64_t *slptep, uint32_t *slpte_level,
710                              bool *reads, bool *writes, uint8_t aw_bits)
711 {
712     dma_addr_t addr = vtd_ce_get_slpt_base(ce);
713     uint32_t level = vtd_ce_get_level(ce);
714     uint32_t offset;
715     uint64_t slpte;
716     uint64_t access_right_check;
717 
718     if (!vtd_iova_range_check(iova, ce, aw_bits)) {
719         error_report_once("%s: detected IOVA overflow (iova=0x%" PRIx64 ")",
720                           __func__, iova);
721         return -VTD_FR_ADDR_BEYOND_MGAW;
722     }
723 
724     /* FIXME: what is the Atomics request here? */
725     access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
726 
727     while (true) {
728         offset = vtd_iova_level_offset(iova, level);
729         slpte = vtd_get_slpte(addr, offset);
730 
731         if (slpte == (uint64_t)-1) {
732             error_report_once("%s: detected read error on DMAR slpte "
733                               "(iova=0x%" PRIx64 ")", __func__, iova);
734             if (level == vtd_ce_get_level(ce)) {
735                 /* Invalid programming of context-entry */
736                 return -VTD_FR_CONTEXT_ENTRY_INV;
737             } else {
738                 return -VTD_FR_PAGING_ENTRY_INV;
739             }
740         }
741         *reads = (*reads) && (slpte & VTD_SL_R);
742         *writes = (*writes) && (slpte & VTD_SL_W);
743         if (!(slpte & access_right_check)) {
744             error_report_once("%s: detected slpte permission error "
745                               "(iova=0x%" PRIx64 ", level=0x%" PRIx32 ", "
746                               "slpte=0x%" PRIx64 ", write=%d)", __func__,
747                               iova, level, slpte, is_write);
748             return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
749         }
750         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
751             error_report_once("%s: detected splte reserve non-zero "
752                               "iova=0x%" PRIx64 ", level=0x%" PRIx32
753                               "slpte=0x%" PRIx64 ")", __func__, iova,
754                               level, slpte);
755             return -VTD_FR_PAGING_ENTRY_RSVD;
756         }
757 
758         if (vtd_is_last_slpte(slpte, level)) {
759             *slptep = slpte;
760             *slpte_level = level;
761             return 0;
762         }
763         addr = vtd_get_slpte_addr(slpte, aw_bits);
764         level--;
765     }
766 }
767 
768 typedef int (*vtd_page_walk_hook)(IOMMUTLBEntry *entry, void *private);
769 
770 /**
771  * Constant information used during page walking
772  *
773  * @hook_fn: hook func to be called when detected page
774  * @private: private data to be passed into hook func
775  * @notify_unmap: whether we should notify invalid entries
776  * @as: VT-d address space of the device
777  * @aw: maximum address width
778  * @domain: domain ID of the page walk
779  */
780 typedef struct {
781     VTDAddressSpace *as;
782     vtd_page_walk_hook hook_fn;
783     void *private;
784     bool notify_unmap;
785     uint8_t aw;
786     uint16_t domain_id;
787 } vtd_page_walk_info;
788 
789 static int vtd_page_walk_one(IOMMUTLBEntry *entry, vtd_page_walk_info *info)
790 {
791     VTDAddressSpace *as = info->as;
792     vtd_page_walk_hook hook_fn = info->hook_fn;
793     void *private = info->private;
794     DMAMap target = {
795         .iova = entry->iova,
796         .size = entry->addr_mask,
797         .translated_addr = entry->translated_addr,
798         .perm = entry->perm,
799     };
800     DMAMap *mapped = iova_tree_find(as->iova_tree, &target);
801 
802     if (entry->perm == IOMMU_NONE && !info->notify_unmap) {
803         trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
804         return 0;
805     }
806 
807     assert(hook_fn);
808 
809     /* Update local IOVA mapped ranges */
810     if (entry->perm) {
811         if (mapped) {
812             /* If it's exactly the same translation, skip */
813             if (!memcmp(mapped, &target, sizeof(target))) {
814                 trace_vtd_page_walk_one_skip_map(entry->iova, entry->addr_mask,
815                                                  entry->translated_addr);
816                 return 0;
817             } else {
818                 /*
819                  * Translation changed.  Normally this should not
820                  * happen, but it can happen when with buggy guest
821                  * OSes.  Note that there will be a small window that
822                  * we don't have map at all.  But that's the best
823                  * effort we can do.  The ideal way to emulate this is
824                  * atomically modify the PTE to follow what has
825                  * changed, but we can't.  One example is that vfio
826                  * driver only has VFIO_IOMMU_[UN]MAP_DMA but no
827                  * interface to modify a mapping (meanwhile it seems
828                  * meaningless to even provide one).  Anyway, let's
829                  * mark this as a TODO in case one day we'll have
830                  * a better solution.
831                  */
832                 IOMMUAccessFlags cache_perm = entry->perm;
833                 int ret;
834 
835                 /* Emulate an UNMAP */
836                 entry->perm = IOMMU_NONE;
837                 trace_vtd_page_walk_one(info->domain_id,
838                                         entry->iova,
839                                         entry->translated_addr,
840                                         entry->addr_mask,
841                                         entry->perm);
842                 ret = hook_fn(entry, private);
843                 if (ret) {
844                     return ret;
845                 }
846                 /* Drop any existing mapping */
847                 iova_tree_remove(as->iova_tree, &target);
848                 /* Recover the correct permission */
849                 entry->perm = cache_perm;
850             }
851         }
852         iova_tree_insert(as->iova_tree, &target);
853     } else {
854         if (!mapped) {
855             /* Skip since we didn't map this range at all */
856             trace_vtd_page_walk_one_skip_unmap(entry->iova, entry->addr_mask);
857             return 0;
858         }
859         iova_tree_remove(as->iova_tree, &target);
860     }
861 
862     trace_vtd_page_walk_one(info->domain_id, entry->iova,
863                             entry->translated_addr, entry->addr_mask,
864                             entry->perm);
865     return hook_fn(entry, private);
866 }
867 
868 /**
869  * vtd_page_walk_level - walk over specific level for IOVA range
870  *
871  * @addr: base GPA addr to start the walk
872  * @start: IOVA range start address
873  * @end: IOVA range end address (start <= addr < end)
874  * @read: whether parent level has read permission
875  * @write: whether parent level has write permission
876  * @info: constant information for the page walk
877  */
878 static int vtd_page_walk_level(dma_addr_t addr, uint64_t start,
879                                uint64_t end, uint32_t level, bool read,
880                                bool write, vtd_page_walk_info *info)
881 {
882     bool read_cur, write_cur, entry_valid;
883     uint32_t offset;
884     uint64_t slpte;
885     uint64_t subpage_size, subpage_mask;
886     IOMMUTLBEntry entry;
887     uint64_t iova = start;
888     uint64_t iova_next;
889     int ret = 0;
890 
891     trace_vtd_page_walk_level(addr, level, start, end);
892 
893     subpage_size = 1ULL << vtd_slpt_level_shift(level);
894     subpage_mask = vtd_slpt_level_page_mask(level);
895 
896     while (iova < end) {
897         iova_next = (iova & subpage_mask) + subpage_size;
898 
899         offset = vtd_iova_level_offset(iova, level);
900         slpte = vtd_get_slpte(addr, offset);
901 
902         if (slpte == (uint64_t)-1) {
903             trace_vtd_page_walk_skip_read(iova, iova_next);
904             goto next;
905         }
906 
907         if (vtd_slpte_nonzero_rsvd(slpte, level)) {
908             trace_vtd_page_walk_skip_reserve(iova, iova_next);
909             goto next;
910         }
911 
912         /* Permissions are stacked with parents' */
913         read_cur = read && (slpte & VTD_SL_R);
914         write_cur = write && (slpte & VTD_SL_W);
915 
916         /*
917          * As long as we have either read/write permission, this is a
918          * valid entry. The rule works for both page entries and page
919          * table entries.
920          */
921         entry_valid = read_cur | write_cur;
922 
923         if (!vtd_is_last_slpte(slpte, level) && entry_valid) {
924             /*
925              * This is a valid PDE (or even bigger than PDE).  We need
926              * to walk one further level.
927              */
928             ret = vtd_page_walk_level(vtd_get_slpte_addr(slpte, info->aw),
929                                       iova, MIN(iova_next, end), level - 1,
930                                       read_cur, write_cur, info);
931         } else {
932             /*
933              * This means we are either:
934              *
935              * (1) the real page entry (either 4K page, or huge page)
936              * (2) the whole range is invalid
937              *
938              * In either case, we send an IOTLB notification down.
939              */
940             entry.target_as = &address_space_memory;
941             entry.iova = iova & subpage_mask;
942             entry.perm = IOMMU_ACCESS_FLAG(read_cur, write_cur);
943             entry.addr_mask = ~subpage_mask;
944             /* NOTE: this is only meaningful if entry_valid == true */
945             entry.translated_addr = vtd_get_slpte_addr(slpte, info->aw);
946             ret = vtd_page_walk_one(&entry, info);
947         }
948 
949         if (ret < 0) {
950             return ret;
951         }
952 
953 next:
954         iova = iova_next;
955     }
956 
957     return 0;
958 }
959 
960 /**
961  * vtd_page_walk - walk specific IOVA range, and call the hook
962  *
963  * @ce: context entry to walk upon
964  * @start: IOVA address to start the walk
965  * @end: IOVA range end address (start <= addr < end)
966  * @info: page walking information struct
967  */
968 static int vtd_page_walk(VTDContextEntry *ce, uint64_t start, uint64_t end,
969                          vtd_page_walk_info *info)
970 {
971     dma_addr_t addr = vtd_ce_get_slpt_base(ce);
972     uint32_t level = vtd_ce_get_level(ce);
973 
974     if (!vtd_iova_range_check(start, ce, info->aw)) {
975         return -VTD_FR_ADDR_BEYOND_MGAW;
976     }
977 
978     if (!vtd_iova_range_check(end, ce, info->aw)) {
979         /* Fix end so that it reaches the maximum */
980         end = vtd_iova_limit(ce, info->aw);
981     }
982 
983     return vtd_page_walk_level(addr, start, end, level, true, true, info);
984 }
985 
986 /* Map a device to its corresponding domain (context-entry) */
987 static int vtd_dev_to_context_entry(IntelIOMMUState *s, uint8_t bus_num,
988                                     uint8_t devfn, VTDContextEntry *ce)
989 {
990     VTDRootEntry re;
991     int ret_fr;
992     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
993 
994     ret_fr = vtd_get_root_entry(s, bus_num, &re);
995     if (ret_fr) {
996         return ret_fr;
997     }
998 
999     if (!vtd_root_entry_present(&re)) {
1000         /* Not error - it's okay we don't have root entry. */
1001         trace_vtd_re_not_present(bus_num);
1002         return -VTD_FR_ROOT_ENTRY_P;
1003     }
1004 
1005     if (re.rsvd || (re.val & VTD_ROOT_ENTRY_RSVD(s->aw_bits))) {
1006         trace_vtd_re_invalid(re.rsvd, re.val);
1007         return -VTD_FR_ROOT_ENTRY_RSVD;
1008     }
1009 
1010     ret_fr = vtd_get_context_entry_from_root(&re, devfn, ce);
1011     if (ret_fr) {
1012         return ret_fr;
1013     }
1014 
1015     if (!vtd_ce_present(ce)) {
1016         /* Not error - it's okay we don't have context entry. */
1017         trace_vtd_ce_not_present(bus_num, devfn);
1018         return -VTD_FR_CONTEXT_ENTRY_P;
1019     }
1020 
1021     if ((ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI) ||
1022                (ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO(s->aw_bits))) {
1023         trace_vtd_ce_invalid(ce->hi, ce->lo);
1024         return -VTD_FR_CONTEXT_ENTRY_RSVD;
1025     }
1026 
1027     /* Check if the programming of context-entry is valid */
1028     if (!vtd_is_level_supported(s, vtd_ce_get_level(ce))) {
1029         trace_vtd_ce_invalid(ce->hi, ce->lo);
1030         return -VTD_FR_CONTEXT_ENTRY_INV;
1031     }
1032 
1033     /* Do translation type check */
1034     if (!vtd_ce_type_check(x86_iommu, ce)) {
1035         trace_vtd_ce_invalid(ce->hi, ce->lo);
1036         return -VTD_FR_CONTEXT_ENTRY_INV;
1037     }
1038 
1039     return 0;
1040 }
1041 
1042 static int vtd_sync_shadow_page_hook(IOMMUTLBEntry *entry,
1043                                      void *private)
1044 {
1045     memory_region_notify_iommu((IOMMUMemoryRegion *)private, 0, *entry);
1046     return 0;
1047 }
1048 
1049 static int vtd_sync_shadow_page_table_range(VTDAddressSpace *vtd_as,
1050                                             VTDContextEntry *ce,
1051                                             hwaddr addr, hwaddr size)
1052 {
1053     IntelIOMMUState *s = vtd_as->iommu_state;
1054     vtd_page_walk_info info = {
1055         .hook_fn = vtd_sync_shadow_page_hook,
1056         .private = (void *)&vtd_as->iommu,
1057         .notify_unmap = true,
1058         .aw = s->aw_bits,
1059         .as = vtd_as,
1060         .domain_id = VTD_CONTEXT_ENTRY_DID(ce->hi),
1061     };
1062 
1063     return vtd_page_walk(ce, addr, addr + size, &info);
1064 }
1065 
1066 static int vtd_sync_shadow_page_table(VTDAddressSpace *vtd_as)
1067 {
1068     int ret;
1069     VTDContextEntry ce;
1070     IOMMUNotifier *n;
1071 
1072     ret = vtd_dev_to_context_entry(vtd_as->iommu_state,
1073                                    pci_bus_num(vtd_as->bus),
1074                                    vtd_as->devfn, &ce);
1075     if (ret) {
1076         if (ret == -VTD_FR_CONTEXT_ENTRY_P) {
1077             /*
1078              * It's a valid scenario to have a context entry that is
1079              * not present.  For example, when a device is removed
1080              * from an existing domain then the context entry will be
1081              * zeroed by the guest before it was put into another
1082              * domain.  When this happens, instead of synchronizing
1083              * the shadow pages we should invalidate all existing
1084              * mappings and notify the backends.
1085              */
1086             IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
1087                 vtd_address_space_unmap(vtd_as, n);
1088             }
1089             ret = 0;
1090         }
1091         return ret;
1092     }
1093 
1094     return vtd_sync_shadow_page_table_range(vtd_as, &ce, 0, UINT64_MAX);
1095 }
1096 
1097 /*
1098  * Fetch translation type for specific device. Returns <0 if error
1099  * happens, otherwise return the shifted type to check against
1100  * VTD_CONTEXT_TT_*.
1101  */
1102 static int vtd_dev_get_trans_type(VTDAddressSpace *as)
1103 {
1104     IntelIOMMUState *s;
1105     VTDContextEntry ce;
1106     int ret;
1107 
1108     s = as->iommu_state;
1109 
1110     ret = vtd_dev_to_context_entry(s, pci_bus_num(as->bus),
1111                                    as->devfn, &ce);
1112     if (ret) {
1113         return ret;
1114     }
1115 
1116     return vtd_ce_get_type(&ce);
1117 }
1118 
1119 static bool vtd_dev_pt_enabled(VTDAddressSpace *as)
1120 {
1121     int ret;
1122 
1123     assert(as);
1124 
1125     ret = vtd_dev_get_trans_type(as);
1126     if (ret < 0) {
1127         /*
1128          * Possibly failed to parse the context entry for some reason
1129          * (e.g., during init, or any guest configuration errors on
1130          * context entries). We should assume PT not enabled for
1131          * safety.
1132          */
1133         return false;
1134     }
1135 
1136     return ret == VTD_CONTEXT_TT_PASS_THROUGH;
1137 }
1138 
1139 /* Return whether the device is using IOMMU translation. */
1140 static bool vtd_switch_address_space(VTDAddressSpace *as)
1141 {
1142     bool use_iommu;
1143     /* Whether we need to take the BQL on our own */
1144     bool take_bql = !qemu_mutex_iothread_locked();
1145 
1146     assert(as);
1147 
1148     use_iommu = as->iommu_state->dmar_enabled & !vtd_dev_pt_enabled(as);
1149 
1150     trace_vtd_switch_address_space(pci_bus_num(as->bus),
1151                                    VTD_PCI_SLOT(as->devfn),
1152                                    VTD_PCI_FUNC(as->devfn),
1153                                    use_iommu);
1154 
1155     /*
1156      * It's possible that we reach here without BQL, e.g., when called
1157      * from vtd_pt_enable_fast_path(). However the memory APIs need
1158      * it. We'd better make sure we have had it already, or, take it.
1159      */
1160     if (take_bql) {
1161         qemu_mutex_lock_iothread();
1162     }
1163 
1164     /* Turn off first then on the other */
1165     if (use_iommu) {
1166         memory_region_set_enabled(&as->sys_alias, false);
1167         memory_region_set_enabled(MEMORY_REGION(&as->iommu), true);
1168     } else {
1169         memory_region_set_enabled(MEMORY_REGION(&as->iommu), false);
1170         memory_region_set_enabled(&as->sys_alias, true);
1171     }
1172 
1173     if (take_bql) {
1174         qemu_mutex_unlock_iothread();
1175     }
1176 
1177     return use_iommu;
1178 }
1179 
1180 static void vtd_switch_address_space_all(IntelIOMMUState *s)
1181 {
1182     GHashTableIter iter;
1183     VTDBus *vtd_bus;
1184     int i;
1185 
1186     g_hash_table_iter_init(&iter, s->vtd_as_by_busptr);
1187     while (g_hash_table_iter_next(&iter, NULL, (void **)&vtd_bus)) {
1188         for (i = 0; i < PCI_DEVFN_MAX; i++) {
1189             if (!vtd_bus->dev_as[i]) {
1190                 continue;
1191             }
1192             vtd_switch_address_space(vtd_bus->dev_as[i]);
1193         }
1194     }
1195 }
1196 
1197 static inline uint16_t vtd_make_source_id(uint8_t bus_num, uint8_t devfn)
1198 {
1199     return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
1200 }
1201 
1202 static const bool vtd_qualified_faults[] = {
1203     [VTD_FR_RESERVED] = false,
1204     [VTD_FR_ROOT_ENTRY_P] = false,
1205     [VTD_FR_CONTEXT_ENTRY_P] = true,
1206     [VTD_FR_CONTEXT_ENTRY_INV] = true,
1207     [VTD_FR_ADDR_BEYOND_MGAW] = true,
1208     [VTD_FR_WRITE] = true,
1209     [VTD_FR_READ] = true,
1210     [VTD_FR_PAGING_ENTRY_INV] = true,
1211     [VTD_FR_ROOT_TABLE_INV] = false,
1212     [VTD_FR_CONTEXT_TABLE_INV] = false,
1213     [VTD_FR_ROOT_ENTRY_RSVD] = false,
1214     [VTD_FR_PAGING_ENTRY_RSVD] = true,
1215     [VTD_FR_CONTEXT_ENTRY_TT] = true,
1216     [VTD_FR_RESERVED_ERR] = false,
1217     [VTD_FR_MAX] = false,
1218 };
1219 
1220 /* To see if a fault condition is "qualified", which is reported to software
1221  * only if the FPD field in the context-entry used to process the faulting
1222  * request is 0.
1223  */
1224 static inline bool vtd_is_qualified_fault(VTDFaultReason fault)
1225 {
1226     return vtd_qualified_faults[fault];
1227 }
1228 
1229 static inline bool vtd_is_interrupt_addr(hwaddr addr)
1230 {
1231     return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <= VTD_INTERRUPT_ADDR_LAST;
1232 }
1233 
1234 static void vtd_pt_enable_fast_path(IntelIOMMUState *s, uint16_t source_id)
1235 {
1236     VTDBus *vtd_bus;
1237     VTDAddressSpace *vtd_as;
1238     bool success = false;
1239 
1240     vtd_bus = vtd_find_as_from_bus_num(s, VTD_SID_TO_BUS(source_id));
1241     if (!vtd_bus) {
1242         goto out;
1243     }
1244 
1245     vtd_as = vtd_bus->dev_as[VTD_SID_TO_DEVFN(source_id)];
1246     if (!vtd_as) {
1247         goto out;
1248     }
1249 
1250     if (vtd_switch_address_space(vtd_as) == false) {
1251         /* We switched off IOMMU region successfully. */
1252         success = true;
1253     }
1254 
1255 out:
1256     trace_vtd_pt_enable_fast_path(source_id, success);
1257 }
1258 
1259 /* Map dev to context-entry then do a paging-structures walk to do a iommu
1260  * translation.
1261  *
1262  * Called from RCU critical section.
1263  *
1264  * @bus_num: The bus number
1265  * @devfn: The devfn, which is the  combined of device and function number
1266  * @is_write: The access is a write operation
1267  * @entry: IOMMUTLBEntry that contain the addr to be translated and result
1268  *
1269  * Returns true if translation is successful, otherwise false.
1270  */
1271 static bool vtd_do_iommu_translate(VTDAddressSpace *vtd_as, PCIBus *bus,
1272                                    uint8_t devfn, hwaddr addr, bool is_write,
1273                                    IOMMUTLBEntry *entry)
1274 {
1275     IntelIOMMUState *s = vtd_as->iommu_state;
1276     VTDContextEntry ce;
1277     uint8_t bus_num = pci_bus_num(bus);
1278     VTDContextCacheEntry *cc_entry;
1279     uint64_t slpte, page_mask;
1280     uint32_t level;
1281     uint16_t source_id = vtd_make_source_id(bus_num, devfn);
1282     int ret_fr;
1283     bool is_fpd_set = false;
1284     bool reads = true;
1285     bool writes = true;
1286     uint8_t access_flags;
1287     VTDIOTLBEntry *iotlb_entry;
1288 
1289     /*
1290      * We have standalone memory region for interrupt addresses, we
1291      * should never receive translation requests in this region.
1292      */
1293     assert(!vtd_is_interrupt_addr(addr));
1294 
1295     vtd_iommu_lock(s);
1296 
1297     cc_entry = &vtd_as->context_cache_entry;
1298 
1299     /* Try to fetch slpte form IOTLB */
1300     iotlb_entry = vtd_lookup_iotlb(s, source_id, addr);
1301     if (iotlb_entry) {
1302         trace_vtd_iotlb_page_hit(source_id, addr, iotlb_entry->slpte,
1303                                  iotlb_entry->domain_id);
1304         slpte = iotlb_entry->slpte;
1305         access_flags = iotlb_entry->access_flags;
1306         page_mask = iotlb_entry->mask;
1307         goto out;
1308     }
1309 
1310     /* Try to fetch context-entry from cache first */
1311     if (cc_entry->context_cache_gen == s->context_cache_gen) {
1312         trace_vtd_iotlb_cc_hit(bus_num, devfn, cc_entry->context_entry.hi,
1313                                cc_entry->context_entry.lo,
1314                                cc_entry->context_cache_gen);
1315         ce = cc_entry->context_entry;
1316         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1317     } else {
1318         ret_fr = vtd_dev_to_context_entry(s, bus_num, devfn, &ce);
1319         is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
1320         if (ret_fr) {
1321             ret_fr = -ret_fr;
1322             if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {
1323                 trace_vtd_fault_disabled();
1324             } else {
1325                 vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);
1326             }
1327             goto error;
1328         }
1329         /* Update context-cache */
1330         trace_vtd_iotlb_cc_update(bus_num, devfn, ce.hi, ce.lo,
1331                                   cc_entry->context_cache_gen,
1332                                   s->context_cache_gen);
1333         cc_entry->context_entry = ce;
1334         cc_entry->context_cache_gen = s->context_cache_gen;
1335     }
1336 
1337     /*
1338      * We don't need to translate for pass-through context entries.
1339      * Also, let's ignore IOTLB caching as well for PT devices.
1340      */
1341     if (vtd_ce_get_type(&ce) == VTD_CONTEXT_TT_PASS_THROUGH) {
1342         entry->iova = addr & VTD_PAGE_MASK_4K;
1343         entry->translated_addr = entry->iova;
1344         entry->addr_mask = ~VTD_PAGE_MASK_4K;
1345         entry->perm = IOMMU_RW;
1346         trace_vtd_translate_pt(source_id, entry->iova);
1347 
1348         /*
1349          * When this happens, it means firstly caching-mode is not
1350          * enabled, and this is the first passthrough translation for
1351          * the device. Let's enable the fast path for passthrough.
1352          *
1353          * When passthrough is disabled again for the device, we can
1354          * capture it via the context entry invalidation, then the
1355          * IOMMU region can be swapped back.
1356          */
1357         vtd_pt_enable_fast_path(s, source_id);
1358         vtd_iommu_unlock(s);
1359         return true;
1360     }
1361 
1362     ret_fr = vtd_iova_to_slpte(&ce, addr, is_write, &slpte, &level,
1363                                &reads, &writes, s->aw_bits);
1364     if (ret_fr) {
1365         ret_fr = -ret_fr;
1366         if (is_fpd_set && vtd_is_qualified_fault(ret_fr)) {
1367             trace_vtd_fault_disabled();
1368         } else {
1369             vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);
1370         }
1371         goto error;
1372     }
1373 
1374     page_mask = vtd_slpt_level_page_mask(level);
1375     access_flags = IOMMU_ACCESS_FLAG(reads, writes);
1376     vtd_update_iotlb(s, source_id, VTD_CONTEXT_ENTRY_DID(ce.hi), addr, slpte,
1377                      access_flags, level);
1378 out:
1379     vtd_iommu_unlock(s);
1380     entry->iova = addr & page_mask;
1381     entry->translated_addr = vtd_get_slpte_addr(slpte, s->aw_bits) & page_mask;
1382     entry->addr_mask = ~page_mask;
1383     entry->perm = access_flags;
1384     return true;
1385 
1386 error:
1387     vtd_iommu_unlock(s);
1388     entry->iova = 0;
1389     entry->translated_addr = 0;
1390     entry->addr_mask = 0;
1391     entry->perm = IOMMU_NONE;
1392     return false;
1393 }
1394 
1395 static void vtd_root_table_setup(IntelIOMMUState *s)
1396 {
1397     s->root = vtd_get_quad_raw(s, DMAR_RTADDR_REG);
1398     s->root_extended = s->root & VTD_RTADDR_RTT;
1399     s->root &= VTD_RTADDR_ADDR_MASK(s->aw_bits);
1400 
1401     trace_vtd_reg_dmar_root(s->root, s->root_extended);
1402 }
1403 
1404 static void vtd_iec_notify_all(IntelIOMMUState *s, bool global,
1405                                uint32_t index, uint32_t mask)
1406 {
1407     x86_iommu_iec_notify_all(X86_IOMMU_DEVICE(s), global, index, mask);
1408 }
1409 
1410 static void vtd_interrupt_remap_table_setup(IntelIOMMUState *s)
1411 {
1412     uint64_t value = 0;
1413     value = vtd_get_quad_raw(s, DMAR_IRTA_REG);
1414     s->intr_size = 1UL << ((value & VTD_IRTA_SIZE_MASK) + 1);
1415     s->intr_root = value & VTD_IRTA_ADDR_MASK(s->aw_bits);
1416     s->intr_eime = value & VTD_IRTA_EIME;
1417 
1418     /* Notify global invalidation */
1419     vtd_iec_notify_all(s, true, 0, 0);
1420 
1421     trace_vtd_reg_ir_root(s->intr_root, s->intr_size);
1422 }
1423 
1424 static void vtd_iommu_replay_all(IntelIOMMUState *s)
1425 {
1426     VTDAddressSpace *vtd_as;
1427 
1428     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1429         vtd_sync_shadow_page_table(vtd_as);
1430     }
1431 }
1432 
1433 static void vtd_context_global_invalidate(IntelIOMMUState *s)
1434 {
1435     trace_vtd_inv_desc_cc_global();
1436     /* Protects context cache */
1437     vtd_iommu_lock(s);
1438     s->context_cache_gen++;
1439     if (s->context_cache_gen == VTD_CONTEXT_CACHE_GEN_MAX) {
1440         vtd_reset_context_cache_locked(s);
1441     }
1442     vtd_iommu_unlock(s);
1443     vtd_address_space_refresh_all(s);
1444     /*
1445      * From VT-d spec 6.5.2.1, a global context entry invalidation
1446      * should be followed by a IOTLB global invalidation, so we should
1447      * be safe even without this. Hoewever, let's replay the region as
1448      * well to be safer, and go back here when we need finer tunes for
1449      * VT-d emulation codes.
1450      */
1451     vtd_iommu_replay_all(s);
1452 }
1453 
1454 /* Do a context-cache device-selective invalidation.
1455  * @func_mask: FM field after shifting
1456  */
1457 static void vtd_context_device_invalidate(IntelIOMMUState *s,
1458                                           uint16_t source_id,
1459                                           uint16_t func_mask)
1460 {
1461     uint16_t mask;
1462     VTDBus *vtd_bus;
1463     VTDAddressSpace *vtd_as;
1464     uint8_t bus_n, devfn;
1465     uint16_t devfn_it;
1466 
1467     trace_vtd_inv_desc_cc_devices(source_id, func_mask);
1468 
1469     switch (func_mask & 3) {
1470     case 0:
1471         mask = 0;   /* No bits in the SID field masked */
1472         break;
1473     case 1:
1474         mask = 4;   /* Mask bit 2 in the SID field */
1475         break;
1476     case 2:
1477         mask = 6;   /* Mask bit 2:1 in the SID field */
1478         break;
1479     case 3:
1480         mask = 7;   /* Mask bit 2:0 in the SID field */
1481         break;
1482     }
1483     mask = ~mask;
1484 
1485     bus_n = VTD_SID_TO_BUS(source_id);
1486     vtd_bus = vtd_find_as_from_bus_num(s, bus_n);
1487     if (vtd_bus) {
1488         devfn = VTD_SID_TO_DEVFN(source_id);
1489         for (devfn_it = 0; devfn_it < PCI_DEVFN_MAX; ++devfn_it) {
1490             vtd_as = vtd_bus->dev_as[devfn_it];
1491             if (vtd_as && ((devfn_it & mask) == (devfn & mask))) {
1492                 trace_vtd_inv_desc_cc_device(bus_n, VTD_PCI_SLOT(devfn_it),
1493                                              VTD_PCI_FUNC(devfn_it));
1494                 vtd_iommu_lock(s);
1495                 vtd_as->context_cache_entry.context_cache_gen = 0;
1496                 vtd_iommu_unlock(s);
1497                 /*
1498                  * Do switch address space when needed, in case if the
1499                  * device passthrough bit is switched.
1500                  */
1501                 vtd_switch_address_space(vtd_as);
1502                 /*
1503                  * So a device is moving out of (or moving into) a
1504                  * domain, resync the shadow page table.
1505                  * This won't bring bad even if we have no such
1506                  * notifier registered - the IOMMU notification
1507                  * framework will skip MAP notifications if that
1508                  * happened.
1509                  */
1510                 vtd_sync_shadow_page_table(vtd_as);
1511             }
1512         }
1513     }
1514 }
1515 
1516 /* Context-cache invalidation
1517  * Returns the Context Actual Invalidation Granularity.
1518  * @val: the content of the CCMD_REG
1519  */
1520 static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t val)
1521 {
1522     uint64_t caig;
1523     uint64_t type = val & VTD_CCMD_CIRG_MASK;
1524 
1525     switch (type) {
1526     case VTD_CCMD_DOMAIN_INVL:
1527         /* Fall through */
1528     case VTD_CCMD_GLOBAL_INVL:
1529         caig = VTD_CCMD_GLOBAL_INVL_A;
1530         vtd_context_global_invalidate(s);
1531         break;
1532 
1533     case VTD_CCMD_DEVICE_INVL:
1534         caig = VTD_CCMD_DEVICE_INVL_A;
1535         vtd_context_device_invalidate(s, VTD_CCMD_SID(val), VTD_CCMD_FM(val));
1536         break;
1537 
1538     default:
1539         error_report_once("%s: invalid context: 0x%" PRIx64,
1540                           __func__, val);
1541         caig = 0;
1542     }
1543     return caig;
1544 }
1545 
1546 static void vtd_iotlb_global_invalidate(IntelIOMMUState *s)
1547 {
1548     trace_vtd_inv_desc_iotlb_global();
1549     vtd_reset_iotlb(s);
1550     vtd_iommu_replay_all(s);
1551 }
1552 
1553 static void vtd_iotlb_domain_invalidate(IntelIOMMUState *s, uint16_t domain_id)
1554 {
1555     VTDContextEntry ce;
1556     VTDAddressSpace *vtd_as;
1557 
1558     trace_vtd_inv_desc_iotlb_domain(domain_id);
1559 
1560     vtd_iommu_lock(s);
1561     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_domain,
1562                                 &domain_id);
1563     vtd_iommu_unlock(s);
1564 
1565     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
1566         if (!vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1567                                       vtd_as->devfn, &ce) &&
1568             domain_id == VTD_CONTEXT_ENTRY_DID(ce.hi)) {
1569             vtd_sync_shadow_page_table(vtd_as);
1570         }
1571     }
1572 }
1573 
1574 static void vtd_iotlb_page_invalidate_notify(IntelIOMMUState *s,
1575                                            uint16_t domain_id, hwaddr addr,
1576                                            uint8_t am)
1577 {
1578     VTDAddressSpace *vtd_as;
1579     VTDContextEntry ce;
1580     int ret;
1581     hwaddr size = (1 << am) * VTD_PAGE_SIZE;
1582 
1583     QLIST_FOREACH(vtd_as, &(s->vtd_as_with_notifiers), next) {
1584         ret = vtd_dev_to_context_entry(s, pci_bus_num(vtd_as->bus),
1585                                        vtd_as->devfn, &ce);
1586         if (!ret && domain_id == VTD_CONTEXT_ENTRY_DID(ce.hi)) {
1587             if (vtd_as_has_map_notifier(vtd_as)) {
1588                 /*
1589                  * As long as we have MAP notifications registered in
1590                  * any of our IOMMU notifiers, we need to sync the
1591                  * shadow page table.
1592                  */
1593                 vtd_sync_shadow_page_table_range(vtd_as, &ce, addr, size);
1594             } else {
1595                 /*
1596                  * For UNMAP-only notifiers, we don't need to walk the
1597                  * page tables.  We just deliver the PSI down to
1598                  * invalidate caches.
1599                  */
1600                 IOMMUTLBEntry entry = {
1601                     .target_as = &address_space_memory,
1602                     .iova = addr,
1603                     .translated_addr = 0,
1604                     .addr_mask = size - 1,
1605                     .perm = IOMMU_NONE,
1606                 };
1607                 memory_region_notify_iommu(&vtd_as->iommu, 0, entry);
1608             }
1609         }
1610     }
1611 }
1612 
1613 static void vtd_iotlb_page_invalidate(IntelIOMMUState *s, uint16_t domain_id,
1614                                       hwaddr addr, uint8_t am)
1615 {
1616     VTDIOTLBPageInvInfo info;
1617 
1618     trace_vtd_inv_desc_iotlb_pages(domain_id, addr, am);
1619 
1620     assert(am <= VTD_MAMV);
1621     info.domain_id = domain_id;
1622     info.addr = addr;
1623     info.mask = ~((1 << am) - 1);
1624     vtd_iommu_lock(s);
1625     g_hash_table_foreach_remove(s->iotlb, vtd_hash_remove_by_page, &info);
1626     vtd_iommu_unlock(s);
1627     vtd_iotlb_page_invalidate_notify(s, domain_id, addr, am);
1628 }
1629 
1630 /* Flush IOTLB
1631  * Returns the IOTLB Actual Invalidation Granularity.
1632  * @val: the content of the IOTLB_REG
1633  */
1634 static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
1635 {
1636     uint64_t iaig;
1637     uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
1638     uint16_t domain_id;
1639     hwaddr addr;
1640     uint8_t am;
1641 
1642     switch (type) {
1643     case VTD_TLB_GLOBAL_FLUSH:
1644         iaig = VTD_TLB_GLOBAL_FLUSH_A;
1645         vtd_iotlb_global_invalidate(s);
1646         break;
1647 
1648     case VTD_TLB_DSI_FLUSH:
1649         domain_id = VTD_TLB_DID(val);
1650         iaig = VTD_TLB_DSI_FLUSH_A;
1651         vtd_iotlb_domain_invalidate(s, domain_id);
1652         break;
1653 
1654     case VTD_TLB_PSI_FLUSH:
1655         domain_id = VTD_TLB_DID(val);
1656         addr = vtd_get_quad_raw(s, DMAR_IVA_REG);
1657         am = VTD_IVA_AM(addr);
1658         addr = VTD_IVA_ADDR(addr);
1659         if (am > VTD_MAMV) {
1660             error_report_once("%s: address mask overflow: 0x%" PRIx64,
1661                               __func__, vtd_get_quad_raw(s, DMAR_IVA_REG));
1662             iaig = 0;
1663             break;
1664         }
1665         iaig = VTD_TLB_PSI_FLUSH_A;
1666         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
1667         break;
1668 
1669     default:
1670         error_report_once("%s: invalid granularity: 0x%" PRIx64,
1671                           __func__, val);
1672         iaig = 0;
1673     }
1674     return iaig;
1675 }
1676 
1677 static void vtd_fetch_inv_desc(IntelIOMMUState *s);
1678 
1679 static inline bool vtd_queued_inv_disable_check(IntelIOMMUState *s)
1680 {
1681     return s->qi_enabled && (s->iq_tail == s->iq_head) &&
1682            (s->iq_last_desc_type == VTD_INV_DESC_WAIT);
1683 }
1684 
1685 static void vtd_handle_gcmd_qie(IntelIOMMUState *s, bool en)
1686 {
1687     uint64_t iqa_val = vtd_get_quad_raw(s, DMAR_IQA_REG);
1688 
1689     trace_vtd_inv_qi_enable(en);
1690 
1691     if (en) {
1692         s->iq = iqa_val & VTD_IQA_IQA_MASK(s->aw_bits);
1693         /* 2^(x+8) entries */
1694         s->iq_size = 1UL << ((iqa_val & VTD_IQA_QS) + 8);
1695         s->qi_enabled = true;
1696         trace_vtd_inv_qi_setup(s->iq, s->iq_size);
1697         /* Ok - report back to driver */
1698         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_QIES);
1699 
1700         if (s->iq_tail != 0) {
1701             /*
1702              * This is a spec violation but Windows guests are known to set up
1703              * Queued Invalidation this way so we allow the write and process
1704              * Invalidation Descriptors right away.
1705              */
1706             trace_vtd_warn_invalid_qi_tail(s->iq_tail);
1707             if (!(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
1708                 vtd_fetch_inv_desc(s);
1709             }
1710         }
1711     } else {
1712         if (vtd_queued_inv_disable_check(s)) {
1713             /* disable Queued Invalidation */
1714             vtd_set_quad_raw(s, DMAR_IQH_REG, 0);
1715             s->iq_head = 0;
1716             s->qi_enabled = false;
1717             /* Ok - report back to driver */
1718             vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_QIES, 0);
1719         } else {
1720             error_report_once("%s: detected improper state when disable QI "
1721                               "(head=0x%x, tail=0x%x, last_type=%d)",
1722                               __func__,
1723                               s->iq_head, s->iq_tail, s->iq_last_desc_type);
1724         }
1725     }
1726 }
1727 
1728 /* Set Root Table Pointer */
1729 static void vtd_handle_gcmd_srtp(IntelIOMMUState *s)
1730 {
1731     vtd_root_table_setup(s);
1732     /* Ok - report back to driver */
1733     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
1734     vtd_reset_caches(s);
1735     vtd_address_space_refresh_all(s);
1736 }
1737 
1738 /* Set Interrupt Remap Table Pointer */
1739 static void vtd_handle_gcmd_sirtp(IntelIOMMUState *s)
1740 {
1741     vtd_interrupt_remap_table_setup(s);
1742     /* Ok - report back to driver */
1743     vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRTPS);
1744 }
1745 
1746 /* Handle Translation Enable/Disable */
1747 static void vtd_handle_gcmd_te(IntelIOMMUState *s, bool en)
1748 {
1749     if (s->dmar_enabled == en) {
1750         return;
1751     }
1752 
1753     trace_vtd_dmar_enable(en);
1754 
1755     if (en) {
1756         s->dmar_enabled = true;
1757         /* Ok - report back to driver */
1758         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
1759     } else {
1760         s->dmar_enabled = false;
1761 
1762         /* Clear the index of Fault Recording Register */
1763         s->next_frcd_reg = 0;
1764         /* Ok - report back to driver */
1765         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
1766     }
1767 
1768     vtd_reset_caches(s);
1769     vtd_address_space_refresh_all(s);
1770 }
1771 
1772 /* Handle Interrupt Remap Enable/Disable */
1773 static void vtd_handle_gcmd_ire(IntelIOMMUState *s, bool en)
1774 {
1775     trace_vtd_ir_enable(en);
1776 
1777     if (en) {
1778         s->intr_enabled = true;
1779         /* Ok - report back to driver */
1780         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_IRES);
1781     } else {
1782         s->intr_enabled = false;
1783         /* Ok - report back to driver */
1784         vtd_set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_IRES, 0);
1785     }
1786 }
1787 
1788 /* Handle write to Global Command Register */
1789 static void vtd_handle_gcmd_write(IntelIOMMUState *s)
1790 {
1791     uint32_t status = vtd_get_long_raw(s, DMAR_GSTS_REG);
1792     uint32_t val = vtd_get_long_raw(s, DMAR_GCMD_REG);
1793     uint32_t changed = status ^ val;
1794 
1795     trace_vtd_reg_write_gcmd(status, val);
1796     if (changed & VTD_GCMD_TE) {
1797         /* Translation enable/disable */
1798         vtd_handle_gcmd_te(s, val & VTD_GCMD_TE);
1799     }
1800     if (val & VTD_GCMD_SRTP) {
1801         /* Set/update the root-table pointer */
1802         vtd_handle_gcmd_srtp(s);
1803     }
1804     if (changed & VTD_GCMD_QIE) {
1805         /* Queued Invalidation Enable */
1806         vtd_handle_gcmd_qie(s, val & VTD_GCMD_QIE);
1807     }
1808     if (val & VTD_GCMD_SIRTP) {
1809         /* Set/update the interrupt remapping root-table pointer */
1810         vtd_handle_gcmd_sirtp(s);
1811     }
1812     if (changed & VTD_GCMD_IRE) {
1813         /* Interrupt remap enable/disable */
1814         vtd_handle_gcmd_ire(s, val & VTD_GCMD_IRE);
1815     }
1816 }
1817 
1818 /* Handle write to Context Command Register */
1819 static void vtd_handle_ccmd_write(IntelIOMMUState *s)
1820 {
1821     uint64_t ret;
1822     uint64_t val = vtd_get_quad_raw(s, DMAR_CCMD_REG);
1823 
1824     /* Context-cache invalidation request */
1825     if (val & VTD_CCMD_ICC) {
1826         if (s->qi_enabled) {
1827             error_report_once("Queued Invalidation enabled, "
1828                               "should not use register-based invalidation");
1829             return;
1830         }
1831         ret = vtd_context_cache_invalidate(s, val);
1832         /* Invalidation completed. Change something to show */
1833         vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
1834         ret = vtd_set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
1835                                       ret);
1836     }
1837 }
1838 
1839 /* Handle write to IOTLB Invalidation Register */
1840 static void vtd_handle_iotlb_write(IntelIOMMUState *s)
1841 {
1842     uint64_t ret;
1843     uint64_t val = vtd_get_quad_raw(s, DMAR_IOTLB_REG);
1844 
1845     /* IOTLB invalidation request */
1846     if (val & VTD_TLB_IVT) {
1847         if (s->qi_enabled) {
1848             error_report_once("Queued Invalidation enabled, "
1849                               "should not use register-based invalidation");
1850             return;
1851         }
1852         ret = vtd_iotlb_flush(s, val);
1853         /* Invalidation completed. Change something to show */
1854         vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
1855         ret = vtd_set_clear_mask_quad(s, DMAR_IOTLB_REG,
1856                                       VTD_TLB_FLUSH_GRANU_MASK_A, ret);
1857     }
1858 }
1859 
1860 /* Fetch an Invalidation Descriptor from the Invalidation Queue */
1861 static bool vtd_get_inv_desc(dma_addr_t base_addr, uint32_t offset,
1862                              VTDInvDesc *inv_desc)
1863 {
1864     dma_addr_t addr = base_addr + offset * sizeof(*inv_desc);
1865     if (dma_memory_read(&address_space_memory, addr, inv_desc,
1866         sizeof(*inv_desc))) {
1867         error_report_once("Read INV DESC failed");
1868         inv_desc->lo = 0;
1869         inv_desc->hi = 0;
1870         return false;
1871     }
1872     inv_desc->lo = le64_to_cpu(inv_desc->lo);
1873     inv_desc->hi = le64_to_cpu(inv_desc->hi);
1874     return true;
1875 }
1876 
1877 static bool vtd_process_wait_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
1878 {
1879     if ((inv_desc->hi & VTD_INV_DESC_WAIT_RSVD_HI) ||
1880         (inv_desc->lo & VTD_INV_DESC_WAIT_RSVD_LO)) {
1881         trace_vtd_inv_desc_wait_invalid(inv_desc->hi, inv_desc->lo);
1882         return false;
1883     }
1884     if (inv_desc->lo & VTD_INV_DESC_WAIT_SW) {
1885         /* Status Write */
1886         uint32_t status_data = (uint32_t)(inv_desc->lo >>
1887                                VTD_INV_DESC_WAIT_DATA_SHIFT);
1888 
1889         assert(!(inv_desc->lo & VTD_INV_DESC_WAIT_IF));
1890 
1891         /* FIXME: need to be masked with HAW? */
1892         dma_addr_t status_addr = inv_desc->hi;
1893         trace_vtd_inv_desc_wait_sw(status_addr, status_data);
1894         status_data = cpu_to_le32(status_data);
1895         if (dma_memory_write(&address_space_memory, status_addr, &status_data,
1896                              sizeof(status_data))) {
1897             trace_vtd_inv_desc_wait_write_fail(inv_desc->hi, inv_desc->lo);
1898             return false;
1899         }
1900     } else if (inv_desc->lo & VTD_INV_DESC_WAIT_IF) {
1901         /* Interrupt flag */
1902         vtd_generate_completion_event(s);
1903     } else {
1904         trace_vtd_inv_desc_wait_invalid(inv_desc->hi, inv_desc->lo);
1905         return false;
1906     }
1907     return true;
1908 }
1909 
1910 static bool vtd_process_context_cache_desc(IntelIOMMUState *s,
1911                                            VTDInvDesc *inv_desc)
1912 {
1913     uint16_t sid, fmask;
1914 
1915     if ((inv_desc->lo & VTD_INV_DESC_CC_RSVD) || inv_desc->hi) {
1916         trace_vtd_inv_desc_cc_invalid(inv_desc->hi, inv_desc->lo);
1917         return false;
1918     }
1919     switch (inv_desc->lo & VTD_INV_DESC_CC_G) {
1920     case VTD_INV_DESC_CC_DOMAIN:
1921         trace_vtd_inv_desc_cc_domain(
1922             (uint16_t)VTD_INV_DESC_CC_DID(inv_desc->lo));
1923         /* Fall through */
1924     case VTD_INV_DESC_CC_GLOBAL:
1925         vtd_context_global_invalidate(s);
1926         break;
1927 
1928     case VTD_INV_DESC_CC_DEVICE:
1929         sid = VTD_INV_DESC_CC_SID(inv_desc->lo);
1930         fmask = VTD_INV_DESC_CC_FM(inv_desc->lo);
1931         vtd_context_device_invalidate(s, sid, fmask);
1932         break;
1933 
1934     default:
1935         trace_vtd_inv_desc_cc_invalid(inv_desc->hi, inv_desc->lo);
1936         return false;
1937     }
1938     return true;
1939 }
1940 
1941 static bool vtd_process_iotlb_desc(IntelIOMMUState *s, VTDInvDesc *inv_desc)
1942 {
1943     uint16_t domain_id;
1944     uint8_t am;
1945     hwaddr addr;
1946 
1947     if ((inv_desc->lo & VTD_INV_DESC_IOTLB_RSVD_LO) ||
1948         (inv_desc->hi & VTD_INV_DESC_IOTLB_RSVD_HI)) {
1949         trace_vtd_inv_desc_iotlb_invalid(inv_desc->hi, inv_desc->lo);
1950         return false;
1951     }
1952 
1953     switch (inv_desc->lo & VTD_INV_DESC_IOTLB_G) {
1954     case VTD_INV_DESC_IOTLB_GLOBAL:
1955         vtd_iotlb_global_invalidate(s);
1956         break;
1957 
1958     case VTD_INV_DESC_IOTLB_DOMAIN:
1959         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
1960         vtd_iotlb_domain_invalidate(s, domain_id);
1961         break;
1962 
1963     case VTD_INV_DESC_IOTLB_PAGE:
1964         domain_id = VTD_INV_DESC_IOTLB_DID(inv_desc->lo);
1965         addr = VTD_INV_DESC_IOTLB_ADDR(inv_desc->hi);
1966         am = VTD_INV_DESC_IOTLB_AM(inv_desc->hi);
1967         if (am > VTD_MAMV) {
1968             trace_vtd_inv_desc_iotlb_invalid(inv_desc->hi, inv_desc->lo);
1969             return false;
1970         }
1971         vtd_iotlb_page_invalidate(s, domain_id, addr, am);
1972         break;
1973 
1974     default:
1975         trace_vtd_inv_desc_iotlb_invalid(inv_desc->hi, inv_desc->lo);
1976         return false;
1977     }
1978     return true;
1979 }
1980 
1981 static bool vtd_process_inv_iec_desc(IntelIOMMUState *s,
1982                                      VTDInvDesc *inv_desc)
1983 {
1984     trace_vtd_inv_desc_iec(inv_desc->iec.granularity,
1985                            inv_desc->iec.index,
1986                            inv_desc->iec.index_mask);
1987 
1988     vtd_iec_notify_all(s, !inv_desc->iec.granularity,
1989                        inv_desc->iec.index,
1990                        inv_desc->iec.index_mask);
1991     return true;
1992 }
1993 
1994 static bool vtd_process_device_iotlb_desc(IntelIOMMUState *s,
1995                                           VTDInvDesc *inv_desc)
1996 {
1997     VTDAddressSpace *vtd_dev_as;
1998     IOMMUTLBEntry entry;
1999     struct VTDBus *vtd_bus;
2000     hwaddr addr;
2001     uint64_t sz;
2002     uint16_t sid;
2003     uint8_t devfn;
2004     bool size;
2005     uint8_t bus_num;
2006 
2007     addr = VTD_INV_DESC_DEVICE_IOTLB_ADDR(inv_desc->hi);
2008     sid = VTD_INV_DESC_DEVICE_IOTLB_SID(inv_desc->lo);
2009     devfn = sid & 0xff;
2010     bus_num = sid >> 8;
2011     size = VTD_INV_DESC_DEVICE_IOTLB_SIZE(inv_desc->hi);
2012 
2013     if ((inv_desc->lo & VTD_INV_DESC_DEVICE_IOTLB_RSVD_LO) ||
2014         (inv_desc->hi & VTD_INV_DESC_DEVICE_IOTLB_RSVD_HI)) {
2015         trace_vtd_inv_desc_iotlb_invalid(inv_desc->hi, inv_desc->lo);
2016         return false;
2017     }
2018 
2019     vtd_bus = vtd_find_as_from_bus_num(s, bus_num);
2020     if (!vtd_bus) {
2021         goto done;
2022     }
2023 
2024     vtd_dev_as = vtd_bus->dev_as[devfn];
2025     if (!vtd_dev_as) {
2026         goto done;
2027     }
2028 
2029     /* According to ATS spec table 2.4:
2030      * S = 0, bits 15:12 = xxxx     range size: 4K
2031      * S = 1, bits 15:12 = xxx0     range size: 8K
2032      * S = 1, bits 15:12 = xx01     range size: 16K
2033      * S = 1, bits 15:12 = x011     range size: 32K
2034      * S = 1, bits 15:12 = 0111     range size: 64K
2035      * ...
2036      */
2037     if (size) {
2038         sz = (VTD_PAGE_SIZE * 2) << cto64(addr >> VTD_PAGE_SHIFT);
2039         addr &= ~(sz - 1);
2040     } else {
2041         sz = VTD_PAGE_SIZE;
2042     }
2043 
2044     entry.target_as = &vtd_dev_as->as;
2045     entry.addr_mask = sz - 1;
2046     entry.iova = addr;
2047     entry.perm = IOMMU_NONE;
2048     entry.translated_addr = 0;
2049     memory_region_notify_iommu(&vtd_dev_as->iommu, 0, entry);
2050 
2051 done:
2052     return true;
2053 }
2054 
2055 static bool vtd_process_inv_desc(IntelIOMMUState *s)
2056 {
2057     VTDInvDesc inv_desc;
2058     uint8_t desc_type;
2059 
2060     trace_vtd_inv_qi_head(s->iq_head);
2061     if (!vtd_get_inv_desc(s->iq, s->iq_head, &inv_desc)) {
2062         s->iq_last_desc_type = VTD_INV_DESC_NONE;
2063         return false;
2064     }
2065     desc_type = inv_desc.lo & VTD_INV_DESC_TYPE;
2066     /* FIXME: should update at first or at last? */
2067     s->iq_last_desc_type = desc_type;
2068 
2069     switch (desc_type) {
2070     case VTD_INV_DESC_CC:
2071         trace_vtd_inv_desc("context-cache", inv_desc.hi, inv_desc.lo);
2072         if (!vtd_process_context_cache_desc(s, &inv_desc)) {
2073             return false;
2074         }
2075         break;
2076 
2077     case VTD_INV_DESC_IOTLB:
2078         trace_vtd_inv_desc("iotlb", inv_desc.hi, inv_desc.lo);
2079         if (!vtd_process_iotlb_desc(s, &inv_desc)) {
2080             return false;
2081         }
2082         break;
2083 
2084     case VTD_INV_DESC_WAIT:
2085         trace_vtd_inv_desc("wait", inv_desc.hi, inv_desc.lo);
2086         if (!vtd_process_wait_desc(s, &inv_desc)) {
2087             return false;
2088         }
2089         break;
2090 
2091     case VTD_INV_DESC_IEC:
2092         trace_vtd_inv_desc("iec", inv_desc.hi, inv_desc.lo);
2093         if (!vtd_process_inv_iec_desc(s, &inv_desc)) {
2094             return false;
2095         }
2096         break;
2097 
2098     case VTD_INV_DESC_DEVICE:
2099         trace_vtd_inv_desc("device", inv_desc.hi, inv_desc.lo);
2100         if (!vtd_process_device_iotlb_desc(s, &inv_desc)) {
2101             return false;
2102         }
2103         break;
2104 
2105     default:
2106         trace_vtd_inv_desc_invalid(inv_desc.hi, inv_desc.lo);
2107         return false;
2108     }
2109     s->iq_head++;
2110     if (s->iq_head == s->iq_size) {
2111         s->iq_head = 0;
2112     }
2113     return true;
2114 }
2115 
2116 /* Try to fetch and process more Invalidation Descriptors */
2117 static void vtd_fetch_inv_desc(IntelIOMMUState *s)
2118 {
2119     trace_vtd_inv_qi_fetch();
2120 
2121     if (s->iq_tail >= s->iq_size) {
2122         /* Detects an invalid Tail pointer */
2123         error_report_once("%s: detected invalid QI tail "
2124                           "(tail=0x%x, size=0x%x)",
2125                           __func__, s->iq_tail, s->iq_size);
2126         vtd_handle_inv_queue_error(s);
2127         return;
2128     }
2129     while (s->iq_head != s->iq_tail) {
2130         if (!vtd_process_inv_desc(s)) {
2131             /* Invalidation Queue Errors */
2132             vtd_handle_inv_queue_error(s);
2133             break;
2134         }
2135         /* Must update the IQH_REG in time */
2136         vtd_set_quad_raw(s, DMAR_IQH_REG,
2137                          (((uint64_t)(s->iq_head)) << VTD_IQH_QH_SHIFT) &
2138                          VTD_IQH_QH_MASK);
2139     }
2140 }
2141 
2142 /* Handle write to Invalidation Queue Tail Register */
2143 static void vtd_handle_iqt_write(IntelIOMMUState *s)
2144 {
2145     uint64_t val = vtd_get_quad_raw(s, DMAR_IQT_REG);
2146 
2147     s->iq_tail = VTD_IQT_QT(val);
2148     trace_vtd_inv_qi_tail(s->iq_tail);
2149 
2150     if (s->qi_enabled && !(vtd_get_long_raw(s, DMAR_FSTS_REG) & VTD_FSTS_IQE)) {
2151         /* Process Invalidation Queue here */
2152         vtd_fetch_inv_desc(s);
2153     }
2154 }
2155 
2156 static void vtd_handle_fsts_write(IntelIOMMUState *s)
2157 {
2158     uint32_t fsts_reg = vtd_get_long_raw(s, DMAR_FSTS_REG);
2159     uint32_t fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2160     uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
2161 
2162     if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
2163         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2164         trace_vtd_fsts_clear_ip();
2165     }
2166     /* FIXME: when IQE is Clear, should we try to fetch some Invalidation
2167      * Descriptors if there are any when Queued Invalidation is enabled?
2168      */
2169 }
2170 
2171 static void vtd_handle_fectl_write(IntelIOMMUState *s)
2172 {
2173     uint32_t fectl_reg;
2174     /* FIXME: when software clears the IM field, check the IP field. But do we
2175      * need to compare the old value and the new value to conclude that
2176      * software clears the IM field? Or just check if the IM field is zero?
2177      */
2178     fectl_reg = vtd_get_long_raw(s, DMAR_FECTL_REG);
2179 
2180     trace_vtd_reg_write_fectl(fectl_reg);
2181 
2182     if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
2183         vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
2184         vtd_set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
2185     }
2186 }
2187 
2188 static void vtd_handle_ics_write(IntelIOMMUState *s)
2189 {
2190     uint32_t ics_reg = vtd_get_long_raw(s, DMAR_ICS_REG);
2191     uint32_t iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2192 
2193     if ((iectl_reg & VTD_IECTL_IP) && !(ics_reg & VTD_ICS_IWC)) {
2194         trace_vtd_reg_ics_clear_ip();
2195         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2196     }
2197 }
2198 
2199 static void vtd_handle_iectl_write(IntelIOMMUState *s)
2200 {
2201     uint32_t iectl_reg;
2202     /* FIXME: when software clears the IM field, check the IP field. But do we
2203      * need to compare the old value and the new value to conclude that
2204      * software clears the IM field? Or just check if the IM field is zero?
2205      */
2206     iectl_reg = vtd_get_long_raw(s, DMAR_IECTL_REG);
2207 
2208     trace_vtd_reg_write_iectl(iectl_reg);
2209 
2210     if ((iectl_reg & VTD_IECTL_IP) && !(iectl_reg & VTD_IECTL_IM)) {
2211         vtd_generate_interrupt(s, DMAR_IEADDR_REG, DMAR_IEDATA_REG);
2212         vtd_set_clear_mask_long(s, DMAR_IECTL_REG, VTD_IECTL_IP, 0);
2213     }
2214 }
2215 
2216 static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
2217 {
2218     IntelIOMMUState *s = opaque;
2219     uint64_t val;
2220 
2221     trace_vtd_reg_read(addr, size);
2222 
2223     if (addr + size > DMAR_REG_SIZE) {
2224         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2225                           " size=0x%u", __func__, addr, size);
2226         return (uint64_t)-1;
2227     }
2228 
2229     switch (addr) {
2230     /* Root Table Address Register, 64-bit */
2231     case DMAR_RTADDR_REG:
2232         if (size == 4) {
2233             val = s->root & ((1ULL << 32) - 1);
2234         } else {
2235             val = s->root;
2236         }
2237         break;
2238 
2239     case DMAR_RTADDR_REG_HI:
2240         assert(size == 4);
2241         val = s->root >> 32;
2242         break;
2243 
2244     /* Invalidation Queue Address Register, 64-bit */
2245     case DMAR_IQA_REG:
2246         val = s->iq | (vtd_get_quad(s, DMAR_IQA_REG) & VTD_IQA_QS);
2247         if (size == 4) {
2248             val = val & ((1ULL << 32) - 1);
2249         }
2250         break;
2251 
2252     case DMAR_IQA_REG_HI:
2253         assert(size == 4);
2254         val = s->iq >> 32;
2255         break;
2256 
2257     default:
2258         if (size == 4) {
2259             val = vtd_get_long(s, addr);
2260         } else {
2261             val = vtd_get_quad(s, addr);
2262         }
2263     }
2264 
2265     return val;
2266 }
2267 
2268 static void vtd_mem_write(void *opaque, hwaddr addr,
2269                           uint64_t val, unsigned size)
2270 {
2271     IntelIOMMUState *s = opaque;
2272 
2273     trace_vtd_reg_write(addr, size, val);
2274 
2275     if (addr + size > DMAR_REG_SIZE) {
2276         error_report_once("%s: MMIO over range: addr=0x%" PRIx64
2277                           " size=0x%u", __func__, addr, size);
2278         return;
2279     }
2280 
2281     switch (addr) {
2282     /* Global Command Register, 32-bit */
2283     case DMAR_GCMD_REG:
2284         vtd_set_long(s, addr, val);
2285         vtd_handle_gcmd_write(s);
2286         break;
2287 
2288     /* Context Command Register, 64-bit */
2289     case DMAR_CCMD_REG:
2290         if (size == 4) {
2291             vtd_set_long(s, addr, val);
2292         } else {
2293             vtd_set_quad(s, addr, val);
2294             vtd_handle_ccmd_write(s);
2295         }
2296         break;
2297 
2298     case DMAR_CCMD_REG_HI:
2299         assert(size == 4);
2300         vtd_set_long(s, addr, val);
2301         vtd_handle_ccmd_write(s);
2302         break;
2303 
2304     /* IOTLB Invalidation Register, 64-bit */
2305     case DMAR_IOTLB_REG:
2306         if (size == 4) {
2307             vtd_set_long(s, addr, val);
2308         } else {
2309             vtd_set_quad(s, addr, val);
2310             vtd_handle_iotlb_write(s);
2311         }
2312         break;
2313 
2314     case DMAR_IOTLB_REG_HI:
2315         assert(size == 4);
2316         vtd_set_long(s, addr, val);
2317         vtd_handle_iotlb_write(s);
2318         break;
2319 
2320     /* Invalidate Address Register, 64-bit */
2321     case DMAR_IVA_REG:
2322         if (size == 4) {
2323             vtd_set_long(s, addr, val);
2324         } else {
2325             vtd_set_quad(s, addr, val);
2326         }
2327         break;
2328 
2329     case DMAR_IVA_REG_HI:
2330         assert(size == 4);
2331         vtd_set_long(s, addr, val);
2332         break;
2333 
2334     /* Fault Status Register, 32-bit */
2335     case DMAR_FSTS_REG:
2336         assert(size == 4);
2337         vtd_set_long(s, addr, val);
2338         vtd_handle_fsts_write(s);
2339         break;
2340 
2341     /* Fault Event Control Register, 32-bit */
2342     case DMAR_FECTL_REG:
2343         assert(size == 4);
2344         vtd_set_long(s, addr, val);
2345         vtd_handle_fectl_write(s);
2346         break;
2347 
2348     /* Fault Event Data Register, 32-bit */
2349     case DMAR_FEDATA_REG:
2350         assert(size == 4);
2351         vtd_set_long(s, addr, val);
2352         break;
2353 
2354     /* Fault Event Address Register, 32-bit */
2355     case DMAR_FEADDR_REG:
2356         if (size == 4) {
2357             vtd_set_long(s, addr, val);
2358         } else {
2359             /*
2360              * While the register is 32-bit only, some guests (Xen...) write to
2361              * it with 64-bit.
2362              */
2363             vtd_set_quad(s, addr, val);
2364         }
2365         break;
2366 
2367     /* Fault Event Upper Address Register, 32-bit */
2368     case DMAR_FEUADDR_REG:
2369         assert(size == 4);
2370         vtd_set_long(s, addr, val);
2371         break;
2372 
2373     /* Protected Memory Enable Register, 32-bit */
2374     case DMAR_PMEN_REG:
2375         assert(size == 4);
2376         vtd_set_long(s, addr, val);
2377         break;
2378 
2379     /* Root Table Address Register, 64-bit */
2380     case DMAR_RTADDR_REG:
2381         if (size == 4) {
2382             vtd_set_long(s, addr, val);
2383         } else {
2384             vtd_set_quad(s, addr, val);
2385         }
2386         break;
2387 
2388     case DMAR_RTADDR_REG_HI:
2389         assert(size == 4);
2390         vtd_set_long(s, addr, val);
2391         break;
2392 
2393     /* Invalidation Queue Tail Register, 64-bit */
2394     case DMAR_IQT_REG:
2395         if (size == 4) {
2396             vtd_set_long(s, addr, val);
2397         } else {
2398             vtd_set_quad(s, addr, val);
2399         }
2400         vtd_handle_iqt_write(s);
2401         break;
2402 
2403     case DMAR_IQT_REG_HI:
2404         assert(size == 4);
2405         vtd_set_long(s, addr, val);
2406         /* 19:63 of IQT_REG is RsvdZ, do nothing here */
2407         break;
2408 
2409     /* Invalidation Queue Address Register, 64-bit */
2410     case DMAR_IQA_REG:
2411         if (size == 4) {
2412             vtd_set_long(s, addr, val);
2413         } else {
2414             vtd_set_quad(s, addr, val);
2415         }
2416         break;
2417 
2418     case DMAR_IQA_REG_HI:
2419         assert(size == 4);
2420         vtd_set_long(s, addr, val);
2421         break;
2422 
2423     /* Invalidation Completion Status Register, 32-bit */
2424     case DMAR_ICS_REG:
2425         assert(size == 4);
2426         vtd_set_long(s, addr, val);
2427         vtd_handle_ics_write(s);
2428         break;
2429 
2430     /* Invalidation Event Control Register, 32-bit */
2431     case DMAR_IECTL_REG:
2432         assert(size == 4);
2433         vtd_set_long(s, addr, val);
2434         vtd_handle_iectl_write(s);
2435         break;
2436 
2437     /* Invalidation Event Data Register, 32-bit */
2438     case DMAR_IEDATA_REG:
2439         assert(size == 4);
2440         vtd_set_long(s, addr, val);
2441         break;
2442 
2443     /* Invalidation Event Address Register, 32-bit */
2444     case DMAR_IEADDR_REG:
2445         assert(size == 4);
2446         vtd_set_long(s, addr, val);
2447         break;
2448 
2449     /* Invalidation Event Upper Address Register, 32-bit */
2450     case DMAR_IEUADDR_REG:
2451         assert(size == 4);
2452         vtd_set_long(s, addr, val);
2453         break;
2454 
2455     /* Fault Recording Registers, 128-bit */
2456     case DMAR_FRCD_REG_0_0:
2457         if (size == 4) {
2458             vtd_set_long(s, addr, val);
2459         } else {
2460             vtd_set_quad(s, addr, val);
2461         }
2462         break;
2463 
2464     case DMAR_FRCD_REG_0_1:
2465         assert(size == 4);
2466         vtd_set_long(s, addr, val);
2467         break;
2468 
2469     case DMAR_FRCD_REG_0_2:
2470         if (size == 4) {
2471             vtd_set_long(s, addr, val);
2472         } else {
2473             vtd_set_quad(s, addr, val);
2474             /* May clear bit 127 (Fault), update PPF */
2475             vtd_update_fsts_ppf(s);
2476         }
2477         break;
2478 
2479     case DMAR_FRCD_REG_0_3:
2480         assert(size == 4);
2481         vtd_set_long(s, addr, val);
2482         /* May clear bit 127 (Fault), update PPF */
2483         vtd_update_fsts_ppf(s);
2484         break;
2485 
2486     case DMAR_IRTA_REG:
2487         if (size == 4) {
2488             vtd_set_long(s, addr, val);
2489         } else {
2490             vtd_set_quad(s, addr, val);
2491         }
2492         break;
2493 
2494     case DMAR_IRTA_REG_HI:
2495         assert(size == 4);
2496         vtd_set_long(s, addr, val);
2497         break;
2498 
2499     default:
2500         if (size == 4) {
2501             vtd_set_long(s, addr, val);
2502         } else {
2503             vtd_set_quad(s, addr, val);
2504         }
2505     }
2506 }
2507 
2508 static IOMMUTLBEntry vtd_iommu_translate(IOMMUMemoryRegion *iommu, hwaddr addr,
2509                                          IOMMUAccessFlags flag, int iommu_idx)
2510 {
2511     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2512     IntelIOMMUState *s = vtd_as->iommu_state;
2513     IOMMUTLBEntry iotlb = {
2514         /* We'll fill in the rest later. */
2515         .target_as = &address_space_memory,
2516     };
2517     bool success;
2518 
2519     if (likely(s->dmar_enabled)) {
2520         success = vtd_do_iommu_translate(vtd_as, vtd_as->bus, vtd_as->devfn,
2521                                          addr, flag & IOMMU_WO, &iotlb);
2522     } else {
2523         /* DMAR disabled, passthrough, use 4k-page*/
2524         iotlb.iova = addr & VTD_PAGE_MASK_4K;
2525         iotlb.translated_addr = addr & VTD_PAGE_MASK_4K;
2526         iotlb.addr_mask = ~VTD_PAGE_MASK_4K;
2527         iotlb.perm = IOMMU_RW;
2528         success = true;
2529     }
2530 
2531     if (likely(success)) {
2532         trace_vtd_dmar_translate(pci_bus_num(vtd_as->bus),
2533                                  VTD_PCI_SLOT(vtd_as->devfn),
2534                                  VTD_PCI_FUNC(vtd_as->devfn),
2535                                  iotlb.iova, iotlb.translated_addr,
2536                                  iotlb.addr_mask);
2537     } else {
2538         error_report_once("%s: detected translation failure "
2539                           "(dev=%02x:%02x:%02x, iova=0x%" PRIx64 ")",
2540                           __func__, pci_bus_num(vtd_as->bus),
2541                           VTD_PCI_SLOT(vtd_as->devfn),
2542                           VTD_PCI_FUNC(vtd_as->devfn),
2543                           iotlb.iova);
2544     }
2545 
2546     return iotlb;
2547 }
2548 
2549 static void vtd_iommu_notify_flag_changed(IOMMUMemoryRegion *iommu,
2550                                           IOMMUNotifierFlag old,
2551                                           IOMMUNotifierFlag new)
2552 {
2553     VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
2554     IntelIOMMUState *s = vtd_as->iommu_state;
2555 
2556     if (!s->caching_mode && new & IOMMU_NOTIFIER_MAP) {
2557         error_report("We need to set caching-mode=1 for intel-iommu to enable "
2558                      "device assignment with IOMMU protection.");
2559         exit(1);
2560     }
2561 
2562     /* Update per-address-space notifier flags */
2563     vtd_as->notifier_flags = new;
2564 
2565     if (old == IOMMU_NOTIFIER_NONE) {
2566         QLIST_INSERT_HEAD(&s->vtd_as_with_notifiers, vtd_as, next);
2567     } else if (new == IOMMU_NOTIFIER_NONE) {
2568         QLIST_REMOVE(vtd_as, next);
2569     }
2570 }
2571 
2572 static int vtd_post_load(void *opaque, int version_id)
2573 {
2574     IntelIOMMUState *iommu = opaque;
2575 
2576     /*
2577      * Memory regions are dynamically turned on/off depending on
2578      * context entry configurations from the guest. After migration,
2579      * we need to make sure the memory regions are still correct.
2580      */
2581     vtd_switch_address_space_all(iommu);
2582 
2583     return 0;
2584 }
2585 
2586 static const VMStateDescription vtd_vmstate = {
2587     .name = "iommu-intel",
2588     .version_id = 1,
2589     .minimum_version_id = 1,
2590     .priority = MIG_PRI_IOMMU,
2591     .post_load = vtd_post_load,
2592     .fields = (VMStateField[]) {
2593         VMSTATE_UINT64(root, IntelIOMMUState),
2594         VMSTATE_UINT64(intr_root, IntelIOMMUState),
2595         VMSTATE_UINT64(iq, IntelIOMMUState),
2596         VMSTATE_UINT32(intr_size, IntelIOMMUState),
2597         VMSTATE_UINT16(iq_head, IntelIOMMUState),
2598         VMSTATE_UINT16(iq_tail, IntelIOMMUState),
2599         VMSTATE_UINT16(iq_size, IntelIOMMUState),
2600         VMSTATE_UINT16(next_frcd_reg, IntelIOMMUState),
2601         VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
2602         VMSTATE_UINT8(iq_last_desc_type, IntelIOMMUState),
2603         VMSTATE_BOOL(root_extended, IntelIOMMUState),
2604         VMSTATE_BOOL(dmar_enabled, IntelIOMMUState),
2605         VMSTATE_BOOL(qi_enabled, IntelIOMMUState),
2606         VMSTATE_BOOL(intr_enabled, IntelIOMMUState),
2607         VMSTATE_BOOL(intr_eime, IntelIOMMUState),
2608         VMSTATE_END_OF_LIST()
2609     }
2610 };
2611 
2612 static const MemoryRegionOps vtd_mem_ops = {
2613     .read = vtd_mem_read,
2614     .write = vtd_mem_write,
2615     .endianness = DEVICE_LITTLE_ENDIAN,
2616     .impl = {
2617         .min_access_size = 4,
2618         .max_access_size = 8,
2619     },
2620     .valid = {
2621         .min_access_size = 4,
2622         .max_access_size = 8,
2623     },
2624 };
2625 
2626 static Property vtd_properties[] = {
2627     DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
2628     DEFINE_PROP_ON_OFF_AUTO("eim", IntelIOMMUState, intr_eim,
2629                             ON_OFF_AUTO_AUTO),
2630     DEFINE_PROP_BOOL("x-buggy-eim", IntelIOMMUState, buggy_eim, false),
2631     DEFINE_PROP_UINT8("x-aw-bits", IntelIOMMUState, aw_bits,
2632                       VTD_HOST_ADDRESS_WIDTH),
2633     DEFINE_PROP_BOOL("caching-mode", IntelIOMMUState, caching_mode, FALSE),
2634     DEFINE_PROP_END_OF_LIST(),
2635 };
2636 
2637 /* Read IRTE entry with specific index */
2638 static int vtd_irte_get(IntelIOMMUState *iommu, uint16_t index,
2639                         VTD_IR_TableEntry *entry, uint16_t sid)
2640 {
2641     static const uint16_t vtd_svt_mask[VTD_SQ_MAX] = \
2642         {0xffff, 0xfffb, 0xfff9, 0xfff8};
2643     dma_addr_t addr = 0x00;
2644     uint16_t mask, source_id;
2645     uint8_t bus, bus_max, bus_min;
2646 
2647     addr = iommu->intr_root + index * sizeof(*entry);
2648     if (dma_memory_read(&address_space_memory, addr, entry,
2649                         sizeof(*entry))) {
2650         error_report_once("%s: read failed: ind=0x%x addr=0x%" PRIx64,
2651                           __func__, index, addr);
2652         return -VTD_FR_IR_ROOT_INVAL;
2653     }
2654 
2655     trace_vtd_ir_irte_get(index, le64_to_cpu(entry->data[1]),
2656                           le64_to_cpu(entry->data[0]));
2657 
2658     if (!entry->irte.present) {
2659         error_report_once("%s: detected non-present IRTE "
2660                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
2661                           __func__, index, le64_to_cpu(entry->data[1]),
2662                           le64_to_cpu(entry->data[0]));
2663         return -VTD_FR_IR_ENTRY_P;
2664     }
2665 
2666     if (entry->irte.__reserved_0 || entry->irte.__reserved_1 ||
2667         entry->irte.__reserved_2) {
2668         error_report_once("%s: detected non-zero reserved IRTE "
2669                           "(index=%u, high=0x%" PRIx64 ", low=0x%" PRIx64 ")",
2670                           __func__, index, le64_to_cpu(entry->data[1]),
2671                           le64_to_cpu(entry->data[0]));
2672         return -VTD_FR_IR_IRTE_RSVD;
2673     }
2674 
2675     if (sid != X86_IOMMU_SID_INVALID) {
2676         /* Validate IRTE SID */
2677         source_id = le32_to_cpu(entry->irte.source_id);
2678         switch (entry->irte.sid_vtype) {
2679         case VTD_SVT_NONE:
2680             break;
2681 
2682         case VTD_SVT_ALL:
2683             mask = vtd_svt_mask[entry->irte.sid_q];
2684             if ((source_id & mask) != (sid & mask)) {
2685                 error_report_once("%s: invalid IRTE SID "
2686                                   "(index=%u, sid=%u, source_id=%u)",
2687                                   __func__, index, sid, source_id);
2688                 return -VTD_FR_IR_SID_ERR;
2689             }
2690             break;
2691 
2692         case VTD_SVT_BUS:
2693             bus_max = source_id >> 8;
2694             bus_min = source_id & 0xff;
2695             bus = sid >> 8;
2696             if (bus > bus_max || bus < bus_min) {
2697                 error_report_once("%s: invalid SVT_BUS "
2698                                   "(index=%u, bus=%u, min=%u, max=%u)",
2699                                   __func__, index, bus, bus_min, bus_max);
2700                 return -VTD_FR_IR_SID_ERR;
2701             }
2702             break;
2703 
2704         default:
2705             error_report_once("%s: detected invalid IRTE SVT "
2706                               "(index=%u, type=%d)", __func__,
2707                               index, entry->irte.sid_vtype);
2708             /* Take this as verification failure. */
2709             return -VTD_FR_IR_SID_ERR;
2710             break;
2711         }
2712     }
2713 
2714     return 0;
2715 }
2716 
2717 /* Fetch IRQ information of specific IR index */
2718 static int vtd_remap_irq_get(IntelIOMMUState *iommu, uint16_t index,
2719                              X86IOMMUIrq *irq, uint16_t sid)
2720 {
2721     VTD_IR_TableEntry irte = {};
2722     int ret = 0;
2723 
2724     ret = vtd_irte_get(iommu, index, &irte, sid);
2725     if (ret) {
2726         return ret;
2727     }
2728 
2729     irq->trigger_mode = irte.irte.trigger_mode;
2730     irq->vector = irte.irte.vector;
2731     irq->delivery_mode = irte.irte.delivery_mode;
2732     irq->dest = le32_to_cpu(irte.irte.dest_id);
2733     if (!iommu->intr_eime) {
2734 #define  VTD_IR_APIC_DEST_MASK         (0xff00ULL)
2735 #define  VTD_IR_APIC_DEST_SHIFT        (8)
2736         irq->dest = (irq->dest & VTD_IR_APIC_DEST_MASK) >>
2737             VTD_IR_APIC_DEST_SHIFT;
2738     }
2739     irq->dest_mode = irte.irte.dest_mode;
2740     irq->redir_hint = irte.irte.redir_hint;
2741 
2742     trace_vtd_ir_remap(index, irq->trigger_mode, irq->vector,
2743                        irq->delivery_mode, irq->dest, irq->dest_mode);
2744 
2745     return 0;
2746 }
2747 
2748 /* Interrupt remapping for MSI/MSI-X entry */
2749 static int vtd_interrupt_remap_msi(IntelIOMMUState *iommu,
2750                                    MSIMessage *origin,
2751                                    MSIMessage *translated,
2752                                    uint16_t sid)
2753 {
2754     int ret = 0;
2755     VTD_IR_MSIAddress addr;
2756     uint16_t index;
2757     X86IOMMUIrq irq = {};
2758 
2759     assert(origin && translated);
2760 
2761     trace_vtd_ir_remap_msi_req(origin->address, origin->data);
2762 
2763     if (!iommu || !iommu->intr_enabled) {
2764         memcpy(translated, origin, sizeof(*origin));
2765         goto out;
2766     }
2767 
2768     if (origin->address & VTD_MSI_ADDR_HI_MASK) {
2769         error_report_once("%s: MSI address high 32 bits non-zero detected: "
2770                           "address=0x%" PRIx64, __func__, origin->address);
2771         return -VTD_FR_IR_REQ_RSVD;
2772     }
2773 
2774     addr.data = origin->address & VTD_MSI_ADDR_LO_MASK;
2775     if (addr.addr.__head != 0xfee) {
2776         error_report_once("%s: MSI address low 32 bit invalid: 0x%" PRIx32,
2777                           __func__, addr.data);
2778         return -VTD_FR_IR_REQ_RSVD;
2779     }
2780 
2781     /* This is compatible mode. */
2782     if (addr.addr.int_mode != VTD_IR_INT_FORMAT_REMAP) {
2783         memcpy(translated, origin, sizeof(*origin));
2784         goto out;
2785     }
2786 
2787     index = addr.addr.index_h << 15 | le16_to_cpu(addr.addr.index_l);
2788 
2789 #define  VTD_IR_MSI_DATA_SUBHANDLE       (0x0000ffff)
2790 #define  VTD_IR_MSI_DATA_RESERVED        (0xffff0000)
2791 
2792     if (addr.addr.sub_valid) {
2793         /* See VT-d spec 5.1.2.2 and 5.1.3 on subhandle */
2794         index += origin->data & VTD_IR_MSI_DATA_SUBHANDLE;
2795     }
2796 
2797     ret = vtd_remap_irq_get(iommu, index, &irq, sid);
2798     if (ret) {
2799         return ret;
2800     }
2801 
2802     if (addr.addr.sub_valid) {
2803         trace_vtd_ir_remap_type("MSI");
2804         if (origin->data & VTD_IR_MSI_DATA_RESERVED) {
2805             error_report_once("%s: invalid IR MSI "
2806                               "(sid=%u, address=0x%" PRIx64
2807                               ", data=0x%" PRIx32 ")",
2808                               __func__, sid, origin->address, origin->data);
2809             return -VTD_FR_IR_REQ_RSVD;
2810         }
2811     } else {
2812         uint8_t vector = origin->data & 0xff;
2813         uint8_t trigger_mode = (origin->data >> MSI_DATA_TRIGGER_SHIFT) & 0x1;
2814 
2815         trace_vtd_ir_remap_type("IOAPIC");
2816         /* IOAPIC entry vector should be aligned with IRTE vector
2817          * (see vt-d spec 5.1.5.1). */
2818         if (vector != irq.vector) {
2819             trace_vtd_warn_ir_vector(sid, index, vector, irq.vector);
2820         }
2821 
2822         /* The Trigger Mode field must match the Trigger Mode in the IRTE.
2823          * (see vt-d spec 5.1.5.1). */
2824         if (trigger_mode != irq.trigger_mode) {
2825             trace_vtd_warn_ir_trigger(sid, index, trigger_mode,
2826                                       irq.trigger_mode);
2827         }
2828     }
2829 
2830     /*
2831      * We'd better keep the last two bits, assuming that guest OS
2832      * might modify it. Keep it does not hurt after all.
2833      */
2834     irq.msi_addr_last_bits = addr.addr.__not_care;
2835 
2836     /* Translate X86IOMMUIrq to MSI message */
2837     x86_iommu_irq_to_msi_message(&irq, translated);
2838 
2839 out:
2840     trace_vtd_ir_remap_msi(origin->address, origin->data,
2841                            translated->address, translated->data);
2842     return 0;
2843 }
2844 
2845 static int vtd_int_remap(X86IOMMUState *iommu, MSIMessage *src,
2846                          MSIMessage *dst, uint16_t sid)
2847 {
2848     return vtd_interrupt_remap_msi(INTEL_IOMMU_DEVICE(iommu),
2849                                    src, dst, sid);
2850 }
2851 
2852 static MemTxResult vtd_mem_ir_read(void *opaque, hwaddr addr,
2853                                    uint64_t *data, unsigned size,
2854                                    MemTxAttrs attrs)
2855 {
2856     return MEMTX_OK;
2857 }
2858 
2859 static MemTxResult vtd_mem_ir_write(void *opaque, hwaddr addr,
2860                                     uint64_t value, unsigned size,
2861                                     MemTxAttrs attrs)
2862 {
2863     int ret = 0;
2864     MSIMessage from = {}, to = {};
2865     uint16_t sid = X86_IOMMU_SID_INVALID;
2866 
2867     from.address = (uint64_t) addr + VTD_INTERRUPT_ADDR_FIRST;
2868     from.data = (uint32_t) value;
2869 
2870     if (!attrs.unspecified) {
2871         /* We have explicit Source ID */
2872         sid = attrs.requester_id;
2873     }
2874 
2875     ret = vtd_interrupt_remap_msi(opaque, &from, &to, sid);
2876     if (ret) {
2877         /* TODO: report error */
2878         /* Drop this interrupt */
2879         return MEMTX_ERROR;
2880     }
2881 
2882     apic_get_class()->send_msi(&to);
2883 
2884     return MEMTX_OK;
2885 }
2886 
2887 static const MemoryRegionOps vtd_mem_ir_ops = {
2888     .read_with_attrs = vtd_mem_ir_read,
2889     .write_with_attrs = vtd_mem_ir_write,
2890     .endianness = DEVICE_LITTLE_ENDIAN,
2891     .impl = {
2892         .min_access_size = 4,
2893         .max_access_size = 4,
2894     },
2895     .valid = {
2896         .min_access_size = 4,
2897         .max_access_size = 4,
2898     },
2899 };
2900 
2901 VTDAddressSpace *vtd_find_add_as(IntelIOMMUState *s, PCIBus *bus, int devfn)
2902 {
2903     uintptr_t key = (uintptr_t)bus;
2904     VTDBus *vtd_bus = g_hash_table_lookup(s->vtd_as_by_busptr, &key);
2905     VTDAddressSpace *vtd_dev_as;
2906     char name[128];
2907 
2908     if (!vtd_bus) {
2909         uintptr_t *new_key = g_malloc(sizeof(*new_key));
2910         *new_key = (uintptr_t)bus;
2911         /* No corresponding free() */
2912         vtd_bus = g_malloc0(sizeof(VTDBus) + sizeof(VTDAddressSpace *) * \
2913                             PCI_DEVFN_MAX);
2914         vtd_bus->bus = bus;
2915         g_hash_table_insert(s->vtd_as_by_busptr, new_key, vtd_bus);
2916     }
2917 
2918     vtd_dev_as = vtd_bus->dev_as[devfn];
2919 
2920     if (!vtd_dev_as) {
2921         snprintf(name, sizeof(name), "intel_iommu_devfn_%d", devfn);
2922         vtd_bus->dev_as[devfn] = vtd_dev_as = g_malloc0(sizeof(VTDAddressSpace));
2923 
2924         vtd_dev_as->bus = bus;
2925         vtd_dev_as->devfn = (uint8_t)devfn;
2926         vtd_dev_as->iommu_state = s;
2927         vtd_dev_as->context_cache_entry.context_cache_gen = 0;
2928         vtd_dev_as->iova_tree = iova_tree_new();
2929 
2930         /*
2931          * Memory region relationships looks like (Address range shows
2932          * only lower 32 bits to make it short in length...):
2933          *
2934          * |-----------------+-------------------+----------|
2935          * | Name            | Address range     | Priority |
2936          * |-----------------+-------------------+----------+
2937          * | vtd_root        | 00000000-ffffffff |        0 |
2938          * |  intel_iommu    | 00000000-ffffffff |        1 |
2939          * |  vtd_sys_alias  | 00000000-ffffffff |        1 |
2940          * |  intel_iommu_ir | fee00000-feefffff |       64 |
2941          * |-----------------+-------------------+----------|
2942          *
2943          * We enable/disable DMAR by switching enablement for
2944          * vtd_sys_alias and intel_iommu regions. IR region is always
2945          * enabled.
2946          */
2947         memory_region_init_iommu(&vtd_dev_as->iommu, sizeof(vtd_dev_as->iommu),
2948                                  TYPE_INTEL_IOMMU_MEMORY_REGION, OBJECT(s),
2949                                  "intel_iommu_dmar",
2950                                  UINT64_MAX);
2951         memory_region_init_alias(&vtd_dev_as->sys_alias, OBJECT(s),
2952                                  "vtd_sys_alias", get_system_memory(),
2953                                  0, memory_region_size(get_system_memory()));
2954         memory_region_init_io(&vtd_dev_as->iommu_ir, OBJECT(s),
2955                               &vtd_mem_ir_ops, s, "intel_iommu_ir",
2956                               VTD_INTERRUPT_ADDR_SIZE);
2957         memory_region_init(&vtd_dev_as->root, OBJECT(s),
2958                            "vtd_root", UINT64_MAX);
2959         memory_region_add_subregion_overlap(&vtd_dev_as->root,
2960                                             VTD_INTERRUPT_ADDR_FIRST,
2961                                             &vtd_dev_as->iommu_ir, 64);
2962         address_space_init(&vtd_dev_as->as, &vtd_dev_as->root, name);
2963         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
2964                                             &vtd_dev_as->sys_alias, 1);
2965         memory_region_add_subregion_overlap(&vtd_dev_as->root, 0,
2966                                             MEMORY_REGION(&vtd_dev_as->iommu),
2967                                             1);
2968         vtd_switch_address_space(vtd_dev_as);
2969     }
2970     return vtd_dev_as;
2971 }
2972 
2973 /* Unmap the whole range in the notifier's scope. */
2974 static void vtd_address_space_unmap(VTDAddressSpace *as, IOMMUNotifier *n)
2975 {
2976     IOMMUTLBEntry entry;
2977     hwaddr size;
2978     hwaddr start = n->start;
2979     hwaddr end = n->end;
2980     IntelIOMMUState *s = as->iommu_state;
2981     DMAMap map;
2982 
2983     /*
2984      * Note: all the codes in this function has a assumption that IOVA
2985      * bits are no more than VTD_MGAW bits (which is restricted by
2986      * VT-d spec), otherwise we need to consider overflow of 64 bits.
2987      */
2988 
2989     if (end > VTD_ADDRESS_SIZE(s->aw_bits)) {
2990         /*
2991          * Don't need to unmap regions that is bigger than the whole
2992          * VT-d supported address space size
2993          */
2994         end = VTD_ADDRESS_SIZE(s->aw_bits);
2995     }
2996 
2997     assert(start <= end);
2998     size = end - start;
2999 
3000     if (ctpop64(size) != 1) {
3001         /*
3002          * This size cannot format a correct mask. Let's enlarge it to
3003          * suite the minimum available mask.
3004          */
3005         int n = 64 - clz64(size);
3006         if (n > s->aw_bits) {
3007             /* should not happen, but in case it happens, limit it */
3008             n = s->aw_bits;
3009         }
3010         size = 1ULL << n;
3011     }
3012 
3013     entry.target_as = &address_space_memory;
3014     /* Adjust iova for the size */
3015     entry.iova = n->start & ~(size - 1);
3016     /* This field is meaningless for unmap */
3017     entry.translated_addr = 0;
3018     entry.perm = IOMMU_NONE;
3019     entry.addr_mask = size - 1;
3020 
3021     trace_vtd_as_unmap_whole(pci_bus_num(as->bus),
3022                              VTD_PCI_SLOT(as->devfn),
3023                              VTD_PCI_FUNC(as->devfn),
3024                              entry.iova, size);
3025 
3026     map.iova = entry.iova;
3027     map.size = entry.addr_mask;
3028     iova_tree_remove(as->iova_tree, &map);
3029 
3030     memory_region_notify_one(n, &entry);
3031 }
3032 
3033 static void vtd_address_space_unmap_all(IntelIOMMUState *s)
3034 {
3035     VTDAddressSpace *vtd_as;
3036     IOMMUNotifier *n;
3037 
3038     QLIST_FOREACH(vtd_as, &s->vtd_as_with_notifiers, next) {
3039         IOMMU_NOTIFIER_FOREACH(n, &vtd_as->iommu) {
3040             vtd_address_space_unmap(vtd_as, n);
3041         }
3042     }
3043 }
3044 
3045 static void vtd_address_space_refresh_all(IntelIOMMUState *s)
3046 {
3047     vtd_address_space_unmap_all(s);
3048     vtd_switch_address_space_all(s);
3049 }
3050 
3051 static int vtd_replay_hook(IOMMUTLBEntry *entry, void *private)
3052 {
3053     memory_region_notify_one((IOMMUNotifier *)private, entry);
3054     return 0;
3055 }
3056 
3057 static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
3058 {
3059     VTDAddressSpace *vtd_as = container_of(iommu_mr, VTDAddressSpace, iommu);
3060     IntelIOMMUState *s = vtd_as->iommu_state;
3061     uint8_t bus_n = pci_bus_num(vtd_as->bus);
3062     VTDContextEntry ce;
3063 
3064     /*
3065      * The replay can be triggered by either a invalidation or a newly
3066      * created entry. No matter what, we release existing mappings
3067      * (it means flushing caches for UNMAP-only registers).
3068      */
3069     vtd_address_space_unmap(vtd_as, n);
3070 
3071     if (vtd_dev_to_context_entry(s, bus_n, vtd_as->devfn, &ce) == 0) {
3072         trace_vtd_replay_ce_valid(bus_n, PCI_SLOT(vtd_as->devfn),
3073                                   PCI_FUNC(vtd_as->devfn),
3074                                   VTD_CONTEXT_ENTRY_DID(ce.hi),
3075                                   ce.hi, ce.lo);
3076         if (vtd_as_has_map_notifier(vtd_as)) {
3077             /* This is required only for MAP typed notifiers */
3078             vtd_page_walk_info info = {
3079                 .hook_fn = vtd_replay_hook,
3080                 .private = (void *)n,
3081                 .notify_unmap = false,
3082                 .aw = s->aw_bits,
3083                 .as = vtd_as,
3084                 .domain_id = VTD_CONTEXT_ENTRY_DID(ce.hi),
3085             };
3086 
3087             vtd_page_walk(&ce, 0, ~0ULL, &info);
3088         }
3089     } else {
3090         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
3091                                     PCI_FUNC(vtd_as->devfn));
3092     }
3093 
3094     return;
3095 }
3096 
3097 /* Do the initialization. It will also be called when reset, so pay
3098  * attention when adding new initialization stuff.
3099  */
3100 static void vtd_init(IntelIOMMUState *s)
3101 {
3102     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3103 
3104     memset(s->csr, 0, DMAR_REG_SIZE);
3105     memset(s->wmask, 0, DMAR_REG_SIZE);
3106     memset(s->w1cmask, 0, DMAR_REG_SIZE);
3107     memset(s->womask, 0, DMAR_REG_SIZE);
3108 
3109     s->root = 0;
3110     s->root_extended = false;
3111     s->dmar_enabled = false;
3112     s->iq_head = 0;
3113     s->iq_tail = 0;
3114     s->iq = 0;
3115     s->iq_size = 0;
3116     s->qi_enabled = false;
3117     s->iq_last_desc_type = VTD_INV_DESC_NONE;
3118     s->next_frcd_reg = 0;
3119     s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND |
3120              VTD_CAP_MAMV | VTD_CAP_PSI | VTD_CAP_SLLPS |
3121              VTD_CAP_SAGAW_39bit | VTD_CAP_MGAW(s->aw_bits);
3122     if (s->aw_bits == VTD_HOST_AW_48BIT) {
3123         s->cap |= VTD_CAP_SAGAW_48bit;
3124     }
3125     s->ecap = VTD_ECAP_QI | VTD_ECAP_IRO;
3126 
3127     /*
3128      * Rsvd field masks for spte
3129      */
3130     vtd_paging_entry_rsvd_field[0] = ~0ULL;
3131     vtd_paging_entry_rsvd_field[1] = VTD_SPTE_PAGE_L1_RSVD_MASK(s->aw_bits);
3132     vtd_paging_entry_rsvd_field[2] = VTD_SPTE_PAGE_L2_RSVD_MASK(s->aw_bits);
3133     vtd_paging_entry_rsvd_field[3] = VTD_SPTE_PAGE_L3_RSVD_MASK(s->aw_bits);
3134     vtd_paging_entry_rsvd_field[4] = VTD_SPTE_PAGE_L4_RSVD_MASK(s->aw_bits);
3135     vtd_paging_entry_rsvd_field[5] = VTD_SPTE_LPAGE_L1_RSVD_MASK(s->aw_bits);
3136     vtd_paging_entry_rsvd_field[6] = VTD_SPTE_LPAGE_L2_RSVD_MASK(s->aw_bits);
3137     vtd_paging_entry_rsvd_field[7] = VTD_SPTE_LPAGE_L3_RSVD_MASK(s->aw_bits);
3138     vtd_paging_entry_rsvd_field[8] = VTD_SPTE_LPAGE_L4_RSVD_MASK(s->aw_bits);
3139 
3140     if (x86_iommu->intr_supported) {
3141         s->ecap |= VTD_ECAP_IR | VTD_ECAP_MHMV;
3142         if (s->intr_eim == ON_OFF_AUTO_ON) {
3143             s->ecap |= VTD_ECAP_EIM;
3144         }
3145         assert(s->intr_eim != ON_OFF_AUTO_AUTO);
3146     }
3147 
3148     if (x86_iommu->dt_supported) {
3149         s->ecap |= VTD_ECAP_DT;
3150     }
3151 
3152     if (x86_iommu->pt_supported) {
3153         s->ecap |= VTD_ECAP_PT;
3154     }
3155 
3156     if (s->caching_mode) {
3157         s->cap |= VTD_CAP_CM;
3158     }
3159 
3160     vtd_reset_caches(s);
3161 
3162     /* Define registers with default values and bit semantics */
3163     vtd_define_long(s, DMAR_VER_REG, 0x10UL, 0, 0);
3164     vtd_define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
3165     vtd_define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
3166     vtd_define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
3167     vtd_define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
3168     vtd_define_long(s, DMAR_GSTS_REG, 0, 0, 0);
3169     vtd_define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffff000ULL, 0);
3170     vtd_define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
3171     vtd_define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
3172 
3173     /* Advanced Fault Logging not supported */
3174     vtd_define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
3175     vtd_define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3176     vtd_define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0);
3177     vtd_define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0);
3178 
3179     /* Treated as RsvdZ when EIM in ECAP_REG is not supported
3180      * vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
3181      */
3182     vtd_define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
3183 
3184     /* Treated as RO for implementations that PLMR and PHMR fields reported
3185      * as Clear in the CAP_REG.
3186      * vtd_define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
3187      */
3188     vtd_define_long(s, DMAR_PMEN_REG, 0, 0, 0);
3189 
3190     vtd_define_quad(s, DMAR_IQH_REG, 0, 0, 0);
3191     vtd_define_quad(s, DMAR_IQT_REG, 0, 0x7fff0ULL, 0);
3192     vtd_define_quad(s, DMAR_IQA_REG, 0, 0xfffffffffffff007ULL, 0);
3193     vtd_define_long(s, DMAR_ICS_REG, 0, 0, 0x1UL);
3194     vtd_define_long(s, DMAR_IECTL_REG, 0x80000000UL, 0x80000000UL, 0);
3195     vtd_define_long(s, DMAR_IEDATA_REG, 0, 0xffffffffUL, 0);
3196     vtd_define_long(s, DMAR_IEADDR_REG, 0, 0xfffffffcUL, 0);
3197     /* Treadted as RsvdZ when EIM in ECAP_REG is not supported */
3198     vtd_define_long(s, DMAR_IEUADDR_REG, 0, 0, 0);
3199 
3200     /* IOTLB registers */
3201     vtd_define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
3202     vtd_define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
3203     vtd_define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
3204 
3205     /* Fault Recording Registers, 128-bit */
3206     vtd_define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
3207     vtd_define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
3208 
3209     /*
3210      * Interrupt remapping registers.
3211      */
3212     vtd_define_quad(s, DMAR_IRTA_REG, 0, 0xfffffffffffff80fULL, 0);
3213 }
3214 
3215 /* Should not reset address_spaces when reset because devices will still use
3216  * the address space they got at first (won't ask the bus again).
3217  */
3218 static void vtd_reset(DeviceState *dev)
3219 {
3220     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3221 
3222     vtd_init(s);
3223     vtd_address_space_refresh_all(s);
3224 }
3225 
3226 static AddressSpace *vtd_host_dma_iommu(PCIBus *bus, void *opaque, int devfn)
3227 {
3228     IntelIOMMUState *s = opaque;
3229     VTDAddressSpace *vtd_as;
3230 
3231     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
3232 
3233     vtd_as = vtd_find_add_as(s, bus, devfn);
3234     return &vtd_as->as;
3235 }
3236 
3237 static bool vtd_decide_config(IntelIOMMUState *s, Error **errp)
3238 {
3239     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(s);
3240 
3241     if (s->intr_eim == ON_OFF_AUTO_ON && !x86_iommu->intr_supported) {
3242         error_setg(errp, "eim=on cannot be selected without intremap=on");
3243         return false;
3244     }
3245 
3246     if (s->intr_eim == ON_OFF_AUTO_AUTO) {
3247         s->intr_eim = (kvm_irqchip_in_kernel() || s->buggy_eim)
3248                       && x86_iommu->intr_supported ?
3249                                               ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
3250     }
3251     if (s->intr_eim == ON_OFF_AUTO_ON && !s->buggy_eim) {
3252         if (!kvm_irqchip_in_kernel()) {
3253             error_setg(errp, "eim=on requires accel=kvm,kernel-irqchip=split");
3254             return false;
3255         }
3256         if (!kvm_enable_x2apic()) {
3257             error_setg(errp, "eim=on requires support on the KVM side"
3258                              "(X2APIC_API, first shipped in v4.7)");
3259             return false;
3260         }
3261     }
3262 
3263     /* Currently only address widths supported are 39 and 48 bits */
3264     if ((s->aw_bits != VTD_HOST_AW_39BIT) &&
3265         (s->aw_bits != VTD_HOST_AW_48BIT)) {
3266         error_setg(errp, "Supported values for x-aw-bits are: %d, %d",
3267                    VTD_HOST_AW_39BIT, VTD_HOST_AW_48BIT);
3268         return false;
3269     }
3270 
3271     return true;
3272 }
3273 
3274 static void vtd_realize(DeviceState *dev, Error **errp)
3275 {
3276     MachineState *ms = MACHINE(qdev_get_machine());
3277     PCMachineState *pcms = PC_MACHINE(ms);
3278     PCIBus *bus = pcms->bus;
3279     IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
3280     X86IOMMUState *x86_iommu = X86_IOMMU_DEVICE(dev);
3281 
3282     x86_iommu->type = TYPE_INTEL;
3283 
3284     if (!vtd_decide_config(s, errp)) {
3285         return;
3286     }
3287 
3288     QLIST_INIT(&s->vtd_as_with_notifiers);
3289     qemu_mutex_init(&s->iommu_lock);
3290     memset(s->vtd_as_by_bus_num, 0, sizeof(s->vtd_as_by_bus_num));
3291     memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
3292                           "intel_iommu", DMAR_REG_SIZE);
3293     sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
3294     /* No corresponding destroy */
3295     s->iotlb = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3296                                      g_free, g_free);
3297     s->vtd_as_by_busptr = g_hash_table_new_full(vtd_uint64_hash, vtd_uint64_equal,
3298                                               g_free, g_free);
3299     vtd_init(s);
3300     sysbus_mmio_map(SYS_BUS_DEVICE(s), 0, Q35_HOST_BRIDGE_IOMMU_ADDR);
3301     pci_setup_iommu(bus, vtd_host_dma_iommu, dev);
3302     /* Pseudo address space under root PCI bus. */
3303     pcms->ioapic_as = vtd_host_dma_iommu(bus, s, Q35_PSEUDO_DEVFN_IOAPIC);
3304 }
3305 
3306 static void vtd_class_init(ObjectClass *klass, void *data)
3307 {
3308     DeviceClass *dc = DEVICE_CLASS(klass);
3309     X86IOMMUClass *x86_class = X86_IOMMU_CLASS(klass);
3310 
3311     dc->reset = vtd_reset;
3312     dc->vmsd = &vtd_vmstate;
3313     dc->props = vtd_properties;
3314     dc->hotpluggable = false;
3315     x86_class->realize = vtd_realize;
3316     x86_class->int_remap = vtd_int_remap;
3317     /* Supported by the pc-q35-* machine types */
3318     dc->user_creatable = true;
3319 }
3320 
3321 static const TypeInfo vtd_info = {
3322     .name          = TYPE_INTEL_IOMMU_DEVICE,
3323     .parent        = TYPE_X86_IOMMU_DEVICE,
3324     .instance_size = sizeof(IntelIOMMUState),
3325     .class_init    = vtd_class_init,
3326 };
3327 
3328 static void vtd_iommu_memory_region_class_init(ObjectClass *klass,
3329                                                      void *data)
3330 {
3331     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_CLASS(klass);
3332 
3333     imrc->translate = vtd_iommu_translate;
3334     imrc->notify_flag_changed = vtd_iommu_notify_flag_changed;
3335     imrc->replay = vtd_iommu_replay;
3336 }
3337 
3338 static const TypeInfo vtd_iommu_memory_region_info = {
3339     .parent = TYPE_IOMMU_MEMORY_REGION,
3340     .name = TYPE_INTEL_IOMMU_MEMORY_REGION,
3341     .class_init = vtd_iommu_memory_region_class_init,
3342 };
3343 
3344 static void vtd_register_types(void)
3345 {
3346     type_register_static(&vtd_info);
3347     type_register_static(&vtd_iommu_memory_region_info);
3348 }
3349 
3350 type_init(vtd_register_types)
3351